Adam, K. Et Al. (AER) - Stock Prices Booms and Expected Capital Gains PDF

American Economic Review 2017, 107(8): 23522408
https://doi.org/10.1257/aer.20140205
Stock Price Booms and Expected Capital Gains
By Klaus Adam, Albert Marcet, and Johannes Beutel*
Investors subjective capital gains expectations are a key element

explaining stock price fluctuations. Survey measures of these expec-
tations display excessive optimism (pessimism) at market peaks
(troughs). We formally reject the hypothesis that this is compatible
with rational expectations. We then incorporate subjective price
beliefs with such properties into a standard asset-pricing model
with rational agents (internal rationality). The model gives rise to
boom-bust cycles that temporarily delink stock prices from funda-
mentals and quantitatively replicates many asset-pricing moments.
In particular, it matches the observed strong positive correlation
between the price dividend ratio and survey return expectations,
which cannot be matched by rational expectations. (JEL D83, D84,
G12, G14)
Bull-markets are born on pessimism, grow on skepticism, mature on

optimism and die on euphoria.
Sir John Templeton, Founder of Templeton Mutual Funds
Following the recent boom and bust cycles in a number of asset markets around the
globe, there exists renewed interest in understanding better the forces c ontributing
*Adam: Department of Economics, University of Mannheim, L7, 3-5, 68131 Mannheim, Germany, and
CEPR (email: adam@uni-mannheim.de); Marcet: Institut dAnlisi Econmica (CSIC), Universitat Autnoma
de Barcelona, Campus Bellaterra, 08193 Barcelona, Spain, and ICREA, MOVE, Barcelona GSE, and CEPR
(email: marcet.albert@gmail.com); Beutel: Deutsche Bundesbank, Wilhelm-Epstein-Strae 14, 60431 Frankfurt
am Main, Germany (email: johannes.beutel@bundesbank.de). This paper replaces an earlier paper titled Booms
and Busts in Asset Prices (Adam and Marcet 2010). Thanks go to Fernando Alvarez, Michele Boldrin, Chryssi
Giannitsarou, Vivien Lewis, Morten Ravn, Ricardo Reis, Mike Woodford, conference participants at the Banque
de France and Chicago Fed Conference on Asset Price Bubbles, ESSIM 2010 in Tarragona, 2017 AEA Meetings,
MONFISPOL network, and seminar participants at Columbia University, University College London, London
School of Economics, Yale University, Harvard University, Northwestern University, Stony Brook, IMF, Mitsui
Conference at Michigan University, University of Chicago, New York University, and London Business School
for helpful comments and suggestions. Research assistance from Jeanine Baumert, Oriol Carreras, Lena Gerko,
Dimitry Matveev, Tongbin Zhang, and especially Sebastian Merkel is greatly appreciated. Klaus Adam thanks the
Bank of Japan for the hospitality offered during early stages of this project. Albert Marcet acknowledges support
from Programa de Excelencia del Banco de Espaa, the Severo Ochoa Programme for Centres of Excellence in
R&D (SEV-2015-0563), Plan Nacional (Ministry of Education), SGR (Generalitat de Catalunya). Klaus Adam and
Albert Marcet acknowledge support from the European Research Council under the EU 7th Framework Programme
(FP/20072013), Starting Grant Agreement 284262 (Adam), and Advanced Grant Agreement 324048 (Marcet).
Klaus Adam and Albert Marcet thank Fondation Banque de France for its support on this project. The paper rep-
resents the authors personal opinions and does not necessarily reflect the views of the Deutsche Bundesbank, the
Eurosystem, or its staff. All errors remain ours. The authors declare that they have no relevant or material financial
interests that relate to the research described in this paper.

Go to https://doi.org/10.1257/aer.20140205 to visit the article page for additional materials and author
disclosure statement(s).
2352
VOL. 107 NO. 8 Adam et al.: Stock price booms and expected capital gains 2353
to the emergence of such drastic asset price movements. This paper argues that
movements in investor optimism and pessimism, as measured by the movements in
investors subjective expectations about future capital gains, are a crucial ingredient
for understanding these fluctuations.
We present an asset-pricing model that incorporates endogenous belief dynamics
about expected capital gains. The model gives rise to sustained stock price booms and
busts and is consistent with the behavior of investors capital gains expectations, as
measured by survey data. The presented modeling approach differs notably from the
standard approach in the consumption-based asset-pricing literature, which proceeds
by assuming that stock price fluctuations are fully efficient. Campbell and Cochrane
(1999) and Bansal and Yaron (2004), for example, present models in which stock
price fluctuations reflect the interaction of investor preferences and stochastic driv-
ing forces in a setting with optimizing investors who hold rational expectations.
We first present empirical evidence casting considerable doubt on the prevailing
view that stock price fluctuations are efficient. Specifically, we show that the rational
expectations (RE) hypothesis gives rise to an important counterfactual prediction
for the behavior of investors return or capital gain expectations.1 This counter-
factual prediction is a model-independent implication of the RE hypothesis, but,
as we explain below, key for understanding stock price volatility and its efficiency
properties.
As previously noted by Fama and French (1988), the empirical behavior of asset
prices implies that rational return expectations correlate negatively with the price
dividend (PD) ratio. Somewhat counterintuitively, the RE hypothesis thus predicts
that investors should be particularly pessimistic about future stock returns in the
early part of the year 2000, when the tech stock boom and the PD ratio of the S&P
500 reached their all-time maximum. As we document, the available survey evi-
dence implies precisely the opposite: all quantitative survey measures of investors
expected return (or capital gain) available for the US economy, unambiguously and
unanimously, correlate positively with the PD ratio. And perhaps not surprisingly,
return expectations reached a temporary maximum rather than a minimum in the
early part of the year 2000, i.e., precisely at the peak of the tech stock boom, a fact
previously shown in Vissing-Jorgensen (2004).
We present formal econometric tests of the null hypothesis that the survey evi-
dence is consistent with RE and demonstrate that the hypothesis of rational return or
capital gain expectations is overwhelmingly rejected by the data. Our tests correct
for small sample bias, account for autocorrelations in the error structure, and are
immune to the presence of differential information on the part of agents and to the
presence of measurement error in survey data. An appealing feature of the tests is
that they also provide clues about why the RE hypothesis fails: the failure arises
because survey expectations and RE covary differently with the PD ratio, a finding
that is useful for guiding the search for alternative and empirically more plausible
expectations models.
The positive comovement of stock prices and survey expectations suggests that
price fluctuations are amplified by overly optimistic beliefs at market peaks and by
1
Since most variation in returns is due to the variation in capital gains, we tend to use both terms interchangeably.
2354 THE AMERICAN ECONOMIC REVIEW August 2017
overly pessimistic beliefs at market troughs. Furthermore, it suggests that inves-

tors capital gains expectations are influenced, at least partly, by the capital gains
observed in the past, in line with evidence presented by Malmendier and Nagel
(2011). Indeed, a simple adaptive updating equation captures the time series behav-
ior of the survey data and its correlation with the PD ratio very well.
Taken together, these observations motivate the construction of an asset-pricing
model in which agents hold subjective beliefs about price outcomes. We do so using
the framework of internal rationality (IR), developed in Adam and Marcet (2011),
which allows us to consider maximizing investors that hold subjective price beliefs
within an otherwise standard Lucas (1978) asset-pricing model.2 Within this frame-
work, agents optimally update beliefs using Bayes law.
With this modification, the Lucas model becomes quantitatively consistent with
important aspects of the data. Using confidence intervals based on the simulated
method of moments (SMM), we find that the model matches key moments describ-
ing the observed volatility of stock prices and the positive correlation between the
PD ratio and subjective return expectations. This is obtained even though we use the
simplest version of the Lucas model with time separable preferences and standard
stochastic driving processes. The same model under RE is very far from explaining
the data and produces, amongst other things, far too little price volatility and the
wrong sign for the correlation between the PD ratio and investors return expectations.
The strong improvement in the models empirical performance arises because
agents attempts to improve their knowledge about price behavior can temporarily
delink asset prices from their fundamental (RE) value and give rise to belief-driven
boom and bust cycles in stock prices. This occurs because with imperfect informa-
tion about the price process, optimal behavior prescribes that agents use past capital
gain observations to learn about the stochastic process governing the behavior of
capital gains; this generates a feedback between capital gain expectations and real-
ized capital gains that can drive booms and busts in stock prices.
Supposein line with the empirical evidencethat agents become more opti-
mistic about future capital gains whenever they are positively surprised by past cap-
ital gains.3 A positive surprise for the capital gains observed in the previous period
then increases optimism about the capital gains associated with investing in the asset
today. If such increased optimism leads to an increase in investors asset demand
and if this demand effect is sufficiently strong, then positive past surprises trigger
further positive surprises today, and thus further increases in optimism tomorrow. As
we show analytically, stock prices in our model do increase with capital gain opti-
mism whenever the substitution effect of increased optimism dominates the wealth
effect of such belief changes. Asset prices in the model then display sustained price
booms, similar to those observed in the data.
After a sequence of sustained increases, countervailing forces come into play that
endogenously dampen the upward price momentum, eventually halt it, and cause a
reversal. Specifically, in a situation where increased optimism about capital gains
2
As is explained in Adam and Marcet (2011), subjective price beliefs are consistent with optimizing behavior
in the presence of lack of common knowledge about agents beliefs and preferences.
3
Such positive surprises may be triggered by fundamental shocks, e.g., a high value for realized dividend
growth.
has led to a stock price boom, stock prices make up for a larger share of agents total
wealth.4 As we show analytically, this causes the wealth effect to become as strong
as (or even stronger than) the substitution effect when expectations about stock price
appreciation are sufficiently high.5 Increases in optimism then cease to cause fur-
ther increases in stock demand and thus stock prices, so that investors capital gains
expectations turn out to be too optimistic relative to the realized outcomes. This
induces downward revision in beliefs, which gives rise to negative price momentum
and an asset price bust.
The previous arguments show how belief dynamics can temporarily delink asset
prices from their fundamental value. Clearly, these price dynamics are inefficient as
they are not justified by innovations to preferences or other fundamentals.
Since we depart from RE, our model requires introducing an explicit assumption
about agents price beliefs. Various elements guide this modeling choice. First, we
choose price beliefs such that there are no black swan-like events, i.e., we insure
that agents have contingency plans for all prices that they actually encounter along
the equilibrium path. Second, we choose the subjective price process such that it
gives rise to capital gain expectations that are consistent with the behavior of survey
expectations. In particular, agents believe the average growth rate of stock prices to
slowly drift over time, which is consistent with the presence of prolonged periods of
price booms that are followed by price busts. Given these beliefs, equilibrium prices
will indeed display prolonged periods of above average and below average growth.
More generally, the present paper shows how the framework of internal ratio-
nality allows studying learning about market behavior in a model of intertemporal
decision making. It thereby improves on shortcomings present in the learning litera-
ture, where agents belief-updating equations and choices are often not derived from
individual maximization and are optimal only in the limit once learning converges to
the RE outcome. We thus provide explicit microfoundations for models of adaptive
learning about market outcomes.
The bulk of the paper considers a representative agent model. This is motivated
by the desire to derive results analytically and to show how a rather small deviation
from the standard paradigm helps reconciling the model with the data. A range of
extensions consider, amongst other things, a heterogeneous agents version and more
elaborate subjective belief structures. These extensions allow replicating additional
data features, e.g., the equity premium.
The remainder of the paper is structured as follows. The next section discusses
the related literature. Section II then shows that the price dividend ratio (PD) ratio
covaries positively with survey measures of investors return expectations and
that this is incompatible with the RE hypothesis. It also shows that the time series
of survey expectations can be captured by a fairly simple belief updating equa-
tion. Section III introduces our asset-pricing model with subjective beliefs. As a
benchmark, Section IV determines the RE equilibrium. Section V introduces a
specific model of subjective price beliefs, which relaxes agents prior about price
4
This occurs because stock prices are high, but also because agents discount other income streams, e.g., wage
income, at a higher rate.
5
With constant relative risk aversion (CRRA) utility, this happens whenever the coefficient of relative risk
aversion is larger than 1.
b ehavior relative to the RE equilibrium beliefs. It also derives the Bayesian updat-
ing equations characterizing the evolution of subjective beliefs over time. After
imposing market clearing in Section VI, we present closed-form solutions for the
PD ratio in SectionVII in the special case of vanishing uncertainty. Using the ana-
lytical solution, we explain how the interaction between belief updating dynamics
and price outcomes can endogenously generate boom and bust dynamics in asset
prices. Section VIII estimates the fully stochastic version of the model using a mix
of calibration and simulated method of moments estimation. It shows that the model
successfully replicates a number of important asset-pricing moments, including the
positive correlation between expected returns and the PD ratio. It also explains how
the model gives rise to a high Sharpe Ratio and to a low volatility for the risk-free
interest rate. Section IX shows that the estimated model can replicate the low fre-
quency movements in the time series of the US postwar PD ratio, as well as the
available time series of survey data. Section X presents a number of robustness
checks and extensions of the basic model. A conclusion briefly summarizes and
discusses potential avenues for future research. Technical material and proofs can
be found in the online Appendix.
I. Related Literature
Following Shillers (1981) seminal observation that stock price volatility cannot
be explained by the volatility of rational dividend expectations, the asset-pricing
literature made considerable progress in explaining stock price behavior. Bansal
and Yaron (2004) and Campbell and Cochrane (1999), for example, developed con-
sumption-based RE models in which price fluctuations result from large and per-
sistent swings in investors stochastic discount factor. Section II shows, however,
that RE models fail to capture the behavior of investors return expectations. This
strongly suggests that RE models fall short of providing a complete explanation of
the sources of stock price volatility.
Attributing stock price fluctuations to sentiment fluctuations or issues of learning
has long had an intuitive appeal. A substantial part of the asset-pricing literature
introduces subjective beliefs to model investor sentiment. The standard approach
resorts to Bayesian RE modeling, which allows for subjective beliefs about funda-
mentals, while keeping the assumption that investors know the equilibrium pricing
function linking stock prices to fundamentals. Following early work by Timmermann
(1993) and Barberis, Shleifer, and Vishny (1998), a substantial literature follows
this approach. It finds that the additional stock price volatility generated from learn-
ing is overall small compared to the gap that exists relative to the data.
Recent work by Barberis et al. (2015), for example, considers a time-separable
utility framework where some investors have rational dividend beliefs while o thers
extrapolate from past dividend observations.6 While this allows one to successfully
6
Barberis et al. (2015, p. 4) claim that agents agree on dividend behavior and that the only difference between
rational agents and extrapolators is that the latter extrapolate future prices from past prices. Yet, as it turns out, their
model is one where some agents extrapolate past dividends while others have rational dividend expectations and
all agents know the equilibrium pricing function, in line with standard Bayesian RE modeling. This can be seen in
the proof of their Proposition 1, which reverse-engineers a process for extrapolators dividend beliefs (ibid., equa-
tions(A16) and (A18)) that differs from the true dividend process (ibid., equation (1)). The proof of the proposition
r eplicate survey behavior, the standard deviation of the PD ratio falls one order of
magnitude short of that observed in the data, so that there are no significant stock
price boom and bust episodes.7
In ongoing work, Hirshleifer, Li, and Yu (2015) and Choi and Mertens (2013)
consider Bayesian RE models with time-nonseparable preferences and investors
who extrapolate past fundamentals. They show how extrapolation of fundamentals
endogenously generates long-run consumption risk and thereby some increase in
asset price volatility. Collin-Dufresne, Johannes, and Lochstoer (2016) show how
learning about fundamentals can give rise to sizable Sharpe ratios, provided agents
have a preference for an early resolution of uncertainty. Standard preference param-
eterizations in this class of models imply, however, that agents are willing to give
up a big part of consumption to resolve consumption risk early (see Epstein, Farhi,
and Strzalecki 2014).
The modeling approach pursued in the present paper differs fundamentally from
the one discussed in the previous paragraphs. The Bayesian RE literature assumes
that agents find it difficult to forecast fundamental shocks (agents hold subjective
dividend beliefs), but that agents can predict perfectly price outcomes conditional
on the history of observed fundamentals (agents know the equilibrium pricing func-
tion mapping the history of dividends into price outcomes). Assuming that agents
know the pricing function provides agents with a substantial amount of information
about market behavior, suggesting that it is of interest to study the effects of relaxing
this informational assumption as we do in the current paper. Our agents entertain a
distribution of prices for given fundamentals, which is nondegenerate and that does
not coincide with the model at hand.8
To show that a key element for understanding stock price volatility is investors
imperfect knowledge about how prices are formed, we make the distinction to the
Bayesian RE literature as stark as possible: we assume that agents find it easy to
predict fundamentals, i.e., assume agents hold RE about dividends, but find it diffi-
cult to predict price behavior, i.e., agents do not know the equilibrium pricing func-
tion. We show that a simple asset-pricing model can then replicate survey data and
generate sufficient volatility for the PD ratio, including occasional boom and bust
episodes. This is achieved in a setting with standard time-separable preferences and
obtained because there is a much stronger propagation of economic disturbances
when agents learn about the equilibrium pricing function: belief changes then affect
stock price behavior and stock prices feed back into belief changes. This allows
movements in prices and beliefs to mutually reinforce each other during price boom
explicitly acknowledges that these subjective dividend beliefs are chosen such that extrapolators price beliefs
(ibid., equations (3) and (4)) are consistent with the equilibrium pricing function: At the same time, in order to
compute the values of the derived parameters that govern their consumption and portfolio decisions, extrapolators
need to be aware of the price equation (A11) (ibid., p. 19). This implies that rational agents and extrapolators
disagree about the dividend process but agree about the equilibrium pricing function, in line with standard Bayesian
RE modeling. It also implies that in their setting, learning from price behavior is observationally equivalent to
learning from dividend behavior, unlike in the setting presented in this paper.
7
Due to the constant absolute risk aversion (CARA) utility setup, the volatility of the PD ratio also asymptot-
ically converges to zero.
8
This is related to work by Angeletos and LaO (2013), who consider a setting in which agents are uncertain
about the price at which they will be able to trade. They show how sentiment shocks can give rise to perfectly
self-fulfilling fluctuations in aggregate outcomes. Sentiment shocks in their setting result from extrinsic uncertainty;
in our setting they are triggered from intrinsic sources of uncertainty.
and bust phases, thereby increasing price volatility. The feedback from market out-
comes into beliefs is absent in a Bayesian RE setting.
The literature on robust control and asset prices (e.g., Cogley and Sargent 2008)
considers settings where investors are uncertain about the process for fundamentals.
In line with Bayesian RE modeling, this literature assumes that investors know the
equilibrium pricing function.
The literature on adaptive learning previously considered deviations from ratio-
nal price expectations using asset-pricing models where investors learn about price
behavior. Marcet and Sargent (1992), for example, study convergence to RE when
agents estimate an incorrect model of stock prices by least squares learning. A range
of papers in the adaptive learning literature argues that learning generates additional
stock price volatility. Bullard and Duffy (2001) and Brock and Hommes (1998), for
example, show that learning dynamics can converge to complicated attractors that
increase asset return volatility, when the RE equilibrium is unstable.9 Lansing (2010)
shows how near-rational bubbles can arise in a model with learning about price
behavior. Branch and Evans (2011) present a model where agents learn about risk
and return and show how it gives rise to bubbles and crashes. Boswijk, Hommes, and
Manzan (2007) estimate a model with fundamentalist and chartist traders whose rel-
ative shares evolve according to an evolutionary performance criterion, showing that
the model can generate a run-up in asset prices and subsequent mean-reversion to fun-
damental values. De Long et al. (1990) show how the pricing effects of positive feed-
back trading survives or even get amplified by the introduction of rational speculators.
The approach used in the present paper differs along several dimensions from
the contributions mentioned in the previous paragraph. First, we compare quantita-
tively the implications of our model with the data, i.e., we match a standard set of
asset-pricing moments capturing stock price volatility and use formal asymptotic dis-
tribution to evaluate the goodness of fit. Second, we compare the model to evidence
obtained from survey data. Third, we present a model that derives investors con-
sumption and stockholding plans from properly specified microfoundations. In par-
ticular, we consider agents who solve an infinite horizon decision problem and hold
a consistent set of beliefs, and we discuss conditions for existence and uniqueness
of optimal plans as well as conditions insuring that the optimal plan has a recursive
representation. The adaptive learning literature often relies on shortcuts that amount
to introducing additional behavioral elements into decision making and postulates
beliefs that become well specified only in the limit, if convergence to RE occurs.10
In prior work, Adam, Marcet, and Nicolini (2016) present a model in which
investors learn about risk-adjusted price growth and show how such a model can
quantitatively replicate a set of standard asset-pricing moments describing stock
price volatility. While replicating stock price volatility and postulating beliefs that
are hard to reject in the light of the existing asset price data and the outcomes gener-
ated by the model, their setup falls short of explaining survey evidence. Specifically,
it counterfactually implies that stock return expectations are constant over time.
Adam, Marcet, and Nicolini (2016) also solve for equilibrium prices under the
assumption that dividend and trading income are a negligible part of total income.
9
Stability under learning dynamics is defined in Marcet and Sargent (1989).
10
See Adam and Marcet (2011, Section 2) for a detailed discussion.
400
350
300
250
PD ratio
200
150
100
50
1950 1960 1970 1980 1990 2000 2010
Figure 1. Quarterly PD Ratio of the S&P 500, 1946:I2012:I
Notes: PD ratio is defined as stock prices over quarterly dividends. Dividends are deseasonalized
by averaging dividend payments over the current and previous three quarters.
We solve the model without this assumption and show that it can play an important
role for the model solution, for example, it gives rise to an endogenous upper bound
for equilibrium prices.11
The experimental and behavioral literature provides further evidence supporting
the presence of subjective price beliefs. Hirota and Sunder (2007) and Asparouhova
et al. (2016), for example, implement the Lucas asset-pricing model in the exper-
imental laboratory and document that there is excess volatility in prices that is
unaccounted for by the rational expectations equilibrium and that likely arises from
participants expectations about future prices. Furthermore, the type of learning
employed in the present model is in line with evidence presented in Malmendier
and Nagel (2011), who show that experienced returns affect beliefs about future
asset returns.12
II. Stock Prices and Stock Price Expectations
This section shows that survey expectations of future stock prices are inconsistent
with the notion that agents hold rational stock price expectations. Indeed, our formal
econometric tests, presented in Sections IIB and IIC, show that stock market inves-
tors display undue optimism about future stock prices when the PD ratio is high and
undue pessimism when the PD ratio is low. Section IID then illustrates how simple
adaptive price predictions quantitatively capture the relationship between survey
expectations and the PD ratio.
11
In line with the approach in the Bayesian RE literature, Adam, Marcet, and Nicolini (2016) impose an exog-
enous upper bound on agents beliefs, a so-called projection facility, so as to insure existence of finite equilibrium
prices.
12
Greenwood and Nagel (2009) show thatin line with this hypothesisyoung mutual fund managers dis-
played trend-chasing behavior over the tech stock boom and bust around the year 2000.
400 0.04
PD ratio
Expected stock returns on own portfolio
Expected stock market returns
350 0.03
Expected return (p.q.)

PD ratio
300 0.02
250 0.01
200 0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Figure 2. Quarterly PD Ratio of the S&P 500 (LHS scale) and Investors Expected
Quarterly Real Returns from the Ubs Gallup Survey (RHS scale), 1998:II2007:IV
A. Survey Expectations and the PD Ratio
As pointed out by Vissing-Jorgensen (2004), Greenwood and Shleifer (2014),

and Adam and Marcet (2010), survey expectations of future returns (or capital
gains) display a positive correlation with the PD ratio, while actual returns (or cap-
ital gains) display a negative correlation.13
Online Appendix A2 documents this fact for a range of surveys and Figure 2
illustrates it using our preferred survey, the UBS Gallup Survey, which is based
on a representative sample of approximately 1,000 US investors who own at least
US$10,000 in financial wealth.14 Figure 2 graphs the US PD ratio (the black line)
together with measures of the cross-sectional average of investors one-year ahead
expected real return.15 Return expectations are expressed in terms of quarterly real
growth rates and the figure depicts two expectations measures: investors expec-
tations about the one year ahead stock market return, as well as their expectations
about the one year ahead returns on their own stock portfolio. These measures behave
very similarly over the period for which they overlap, but the latter is available
for a longer time period. Figure 2 reveals that there is a strong positive correlation
between the PD ratio and expected returns. The correlation between the expected
own portfolio returns and the PD ratio is +0.70 and even higher for expected stock
market returns (+0.79). Moreover, investors return expectations were highest at
the beginning of the year 2000, which is precisely the year the PD ratio reached its
13
A related observation is that return forecast errors implied by survey data can be predicted using the PD ratio:
see Bacchetta, Mertens, and Van Wincoop (2009).
14
About 40 percent of respondents own more than US$100,000 in financial wealth. As is documented in online
Appendix A2, this subgroup does not behave differently.
15
To be consistent with the asset-pricing model presented in later sections, we report expectations of real
returns. The nominal return expectations from the survey have been transformed into real returns using inflation
forecasts from the Survey of Professional Forecasters. Results are robust to using instead the Michigan Survey
inflation forecast.
peak during the tech stock boom. Investors then expected annualized real returns of
around 13percent from stock investments, while the subsequently realized returns
turned out to be particularly dismal. Conversely, investors were most pessimistic in
the year 2003 when the PD ratio reached its bottom, expecting then annualized real
returns of below 4 percent.
This evidence suggests that survey data are incompatible with rational expec-
tations and that stock prices seem to play a role in the formation of expectations
about stock returns. Yet, evidence based on comparing two correlations can only
be suggestive, as it is subject to several econometric shortcomings. For example, if
investors possess information that is not observed by the econometrician, as might
be considered likely, then the correlation between the fully rational return forecasts
and the PD ratio will differ from the correlation between realized returns and the
PD ratio. The same holds true if survey expectations are measured with error, as
one can reasonably expect. Furthermore, results in Stambaugh (1999) imply that
with the PD ratio being such a persistent process, there is considerable small sample
bias in these correlations, given the relatively short time spans over which investor
expectations can be tracked. Finally, a highly serially correlated predictor variable
(PD ratio), whose innovations are correlated with the variable that is to be predicted
(future returns), gives rise to spurious regression and thus spurious correlation prob-
lems (see Ferson, Sarkissian, and Simin 2003 and Campbell and Yogo 2006). There
also does not exist a standard nonparametric approach allowing to correct for these
small sample issues when comparing correlations.16 Comparisons involving cor-
relations are thus insufficient for rejecting the hypothesis that survey expectations
are consistent with RE. To deal with these concerns, the next sections construct
formal econometric tests that take these concerns fully into account.
B. RE Test with Small Sample Adjustments
This section develops a RE test that takes into account the concerns expressed
in the previous section. While the present section emphasizes the derivation of ana-
lytical results, Section IIC provides further tests that rely entirely on Monte Carlo
simulation.
Let E t denote agents subjective expectations operator based on informa-
tion up to time t, which can differ from the rational expectations operator Et.
Let Rt,t+Ndenote the real cumulative stock returns between period tand t+N and
let Nt =Et Rt,t+N
+ Nt denote the (potentially noisy) observation of expected
returns, as obtained, for example, from survey data, where Nt is measurement
error.17
P
Let us linearly project the random variable Et Rt,t+N on __
Dt to define
t
P
=a N+c N___t+utN ,
Et Rt,t+N
(1)
Dt
16
Any test must take into account the joint distribution of the correlation estimates in order to make statistically
valid statements.
17
Since the Shiller survey reports expectations about capital gains instead of returns, Rt,t+N
denotes the real
growth rate of stock prices between periods tand t+Nwhen using the Shiller survey.
where
E(xtutN )=0,
(2)
(1,Pt /Dt). The operator E

for x t= denotes the objective expectation for the true
data-generating process, whatever is the process for agents expectations. The pro-
jection residual u tN captures variations in agents actual expectations that cannot be
linearly attributed to the price-dividend ratio.18 It summarizes all other information
that agents believe to be useful in predicting Rt,t+N.19
Due to the potential presence of measurement error, one cannot directly estimate
equation (1), but given the observed return expectations Nt , one can write the fol-
lowing regression equation:
P
Nt =a N+c N___t+utN + Nt .
(3)
Dt
Assuming that the measurement error

Nt is orthogonal to the current PD ratio,20 we
have the orthogonality condition
E[xt (utN + Nt )] = 0.
(4)
Let cNT denote the ordinary least squares (OLS) estimator of c Nin equation (3),
given a sample of length T .21
We can specify an additional regression equation like equation (3), but with
actual returns as dependent variable,
P
Rt,t+N =a N+c N___t+u Nt ,
(5)
Dt
where
E[ xtu Nt ] = 0.
(6)
Let c NT denote the OLS estimate of c Nwith T observations.

The reader can probably guess that the regression estimates are useful here
because under the hypothesis of RE we have c N=c N, so that the estimates
cT and c T are both consistent estimates of the same parameter, letting the prediction
N N
error Nt =Rt, t+NEt(Rt,

t+N), where Et is taken with respect to the objective dis-
tribution and the information available to investors at t, we have u N =u N+ Nt .
This gives rise to the following test.
18
The residual utN is likely to be correlated with current and past observables (other than the PD ratio) and thus
serially correlated.
19
The projection in equation (1) and the error are well defined as long as agents expectations E t Rt,t+N
and Pt /Dtare stationary and have bounded second moments.
20
We allow Nt to be serially correlated and correlated with equilibrium variables other than PDt.
21
Although the residuals u tN and the measurement errors Nt are likely to be serially correlated, the OLS esti-
mate is consistent.
Proposition 1: Assume the process { , Nt ,Pt /Dt}is station-

Rt,t+N

ary and ergodic, all moments are such that asymptotic distributions exist,22
E(t )=E(t Pt /Dt )=0, and P
t /Dtis part of agents time t information set. Then
(i) Under the null hypothesis of rational expectations
__ c NT cNT
(7) ______
T N(0,1) in distribution as T,
cc
is a consistent estimate of var(cNcN).23

where 2cc
(ii) Suppose in addition that
P Pt1
___t=
(8) PD(1)+____ t
+ PD
Dt Dt1
for (1,1) , where (utN + Nt ,u Nt , PD

t+1
) is normally distributed, i.i.d.,
with mean 0, and E( Nt PD
t+1

)=0 . Under the null hypothesis of rational
expectations, the small sample bias of cT cNT in the test-statistic (7) is
N

___________ t+1
cov( PD , Nt )
(9)
E(c NT cNT )=
E(T ),
var( t )
PD
where E(T ) is the small sample bias in the estimation of for a sample
of length T .
The proof of Proposition 1 can be found in online Appendix A3. It treats equa-
tions (3)(6) as a seemingly unrelated regression system and uses the fact that under
rational expectations one has c N=c N. Part (ii) of the proposition follows from
results in Stambaugh (1999).
Part (i) of Proposition 1 uses minimal assumptions to obtain an asymptotically
valid result. Essentially, all that is needed is stationarity of the observables and
orthogonality of the measurement error. The test is asymptotically robust to serial
correlation and heteroskedasticity of the error terms.
Part (ii) of Proposition 1 deals with small sample bias in the test statistic. Since
is close to 1, we have E(T )<0; and since future stock price increases are
likely to be correlated with future surprises to returns, i.e., c ov( PD t+1, t )>0 , we
N
tend to get E (c T cT )<0in small samples, even if in fact c =c .

N N N N
22
More precisely, we assume: (i) bounded second moments of (Rt,t+N
, Nt ,Pt /Dt ) ; (ii) var(Pt /Dt )>0; and
k= ([ ut + t ] [ u tk )
+tk
]

u N + Nt utkN
Sw =
E tN
N xtx tk
<.
N
+ Ntk

23
Equation (A5) in the proof of Proposition 1 provides an explicit expression for a consistent variance estimator.
The fraction on the right-hand side of the bias expression in equation (9) can be
estimated from observables using the calculated errors from equations (3), (5), and
(8) and the fact that under the null
t+1, t )=cov( t+1

cov( PD N PD
,u Nt )cov( PD
t+1
,utN + Nt ).
The bias E (T )in equation (9) is approximately given by E (T )

____
T
1+3

(see Marriott and Pope 1954). Since the true bias is a nonlinear function
of and the analytical linear approximation less precise in the relevant range of
close to 1,24 we compute the bias E (T )using Monte Carlo integration.25
Interestingly, our RE test c N=c Nis less prone to small sample bias than tests
for the significance of the individual regression coefficients (c N = 0and c N = 0).
This follows from the proof of Proposition 1, which shows that
t+1
cov( PD , Nt )
E(c NT c N )=E(cNT c N )+___________
E(T ).
var( t )
PD
The previous equation implies that the small sample bias present in the individual
estimate of c N, i.e., E
(cNT c N ) cancels in the numerator of test statistic (7) when
testing for c N=c N.
The test outcomes associated with Proposition 1 are reported in Table 1A.26
The table reports the bias-corrected point estimates of c N and c N, as well as the
bias-corrected p-values for the test based on Proposition 1.27 We use the UBS, CFO,
and Shiller surveys and consider various ways for extracting expectations from these
surveys.28 The point estimates always satisfy c N>0 and c N<0. The difference
between the two estimates is statistically significant at the 1 percent level in all
cases, except when using the survey median from the CFO survey, where p-values
are around 3 to 5 percent.29
Overall, the test results in Table 1A provide strong evidence against the notion
that survey expectations are compatible with rational expectations. Table 1A also
shows that agents are overly optimistic when the P Dratio is high and overly pessi-
mistic when the P Dratio is low. This suggests that current prices have an excessive
role in influencing current return expectations. Clearly, if the asset price and survey
data were generated by a rational expectations modelsay, the models of Campbell
and Cochrane (1999) or Bansal and Yaron (2004)the tests in Table 1A would have
been accepted.
24
See, for instance, MacKinnon and Smith (1998, Figure 1).
25
Given the estimated values of
PD,, 2we simulate 10,000 realizations of P
PD Dof length T , compute for
each realization, average over realizations to obtain an approximation for E(T ), and compute the bias correction
accordingly.
26
We used 4 lags for Newey West estimator and we checked that results are robust to increasing the lag length
up to 12 lags. For each considered survey, we use data on actual returns (or excess returns, or price growth) for the
same time period for which survey data are available when computing the p-values.
__ cN cN E(cN cN )
27
The p-values are computed using the bias corrected test statistic _____________
T
T T
cc

T T

.
28
See online Appendix A1 for information on the data sources.
29
We conjecture that the CFO provides less significant results because the sample starts in 2000:III, and thus
does not include the upswing of the tech boom period, unlike the UBS sample. As a result, the CFO sample period
displays less mean reversion in prices, which accounts for the fact that the estimates of care less negative and less
significant.
Table 1ARational Expectations Test, H 0: c=c

(Stock Returns, Small Sample Bias Correction from Proposition 1)
Survey average Survey median

bias 103 p-value bias 103 p-value
c 103 c 103 E(c c) H
0 : c=c c 103 c 103 E(c c) H0 : c=c
Panel A. S&P 500, real returns
UBS, >100k, 1 yr, SPF 0.58 2.46 0.432 0.0000 0.48 2.49 0.415 0.0000
UBS, >100k, 1 yr, Michigan 0.57 2.46 0.452 0.0000 0.47 2.49 0.413 0.0000
UBS, all, 1 yr, SPF 0.57 2.46 0.424 0.0000 0.49 2.49 0.401 0.0000
UBS, all, 1 yr, Michigan 0.56 2.46 0.442 0.0000 0.48 2.49 0.433 0.0000
CFO, 1 yr, SPF 0.20 1.67 0.222 0.0011 0.25 1.37 0.325 0.0471
CFO, 1 yr, Michigan 0.27 1.67 0.200 0.0006 0.34 1.37 0.313 0.0362
Panel B. Dow Jones, real price growth

Shiller, 1 yr, SPF 0.26 1.22 0.235 0.0011 0.24 1.20 0.265 0.0015
Shiller, 1 yr, Michigan 0.33 1.22 0.232 0.0006 0.31 1.20 0.238 0.0007
Shiller, 10 yrs, SPF 4.73 7.25 1.367 0.0000 6.15 7.24 1.440 0.0000
Shiller, 10 yrs, Michigan 4.24 7.25 1.423 0.0000 5.65 7.24 1.462 0.0000
Notes: The table reports p-values for the rational expectations test from Section IIB for different survey sources and
different ways to extract expectations from the surveys. Estimates and p-values are bias corrected as described in
Section IIB and Proposition 1. c is the estimate of c N in equation (3) and c the estimate of c N in equation (5). The
column labeled Bias reports the small sample bias of c cas implied by Proposition 1. The UBS and CFO surveys
report return expectations for the S&P 500, the Shiller surveys report capital gain expectations for the Dow Jones
Index. In the regressions, R
t,t+N denotes returns, except for the Shiller survey, where it denotes capital gains. The
columns labeled survey average compute expectations using the cross-sectional average of return survey expecta-
tions; the columns labeled survey median use the median survey expectation. In the first column, SPF and Michigan
refer to different approaches to compute real expected returns, with the former using inflation expectations from the
Survey of Professional Forecasters (SPF) and the latter using the Michigan survey; 1 yr and 10 yr refer to forecast
horizons of 1 and 10 years, respectively; UBS, >100k indicates a restricted sample using only UBS survey partici-
pants with more than US$100,000 in financial wealth; UBS, all indicates the use of all survey participants.
C. Additional Small Sample Adjustments
The closed-form expressions for the small sample bias derived in the previous
section are useful for understanding the nature of the bias. At the same time, they fall
short of completely addressing small sample issues. In particular, the result stated in
Proposition 1(ii) relies on assuming that the regression residuals utN + Nt and u Nt in
equations (3) and (5) are i.i.d. Yet, this is unlikely to hold in our application.
Consider first the residual u Nt
. Under the null hypothesis of RE, we
have u Nt =utN + Nt , where Nt Rt,t+N EtRt,t+N
is a prediction error from the
true data-generating process. For short prediction horizons (N=1), Nt is indeed
serially uncorrelated, but for longer horizons (N>1), the residuals Nt denote fore-
cast errors for overlapping prediction horizons. Since we use quarterly data and
prediction horizons between one and ten years, serial correlation of u Nt is a relevant
concern.
Next, consider the residual utN . Under the null hypothesis, it captures information
that agents use to predict future returns over and above the PD ratio. The variables
capturing such information can be expected to be themselves serially correlated.30
30
For example, in a model with capital accumulation, u tN is a function of the capital stock, which is highly
serially correlated.
Indeed, once we take into account serial correlation below, we find a quarterly per-
sistence for utN between 0.77 and 0.93.
An additional concern with the small sample result in Proposition 1 is that the
variance ccis usually underestimated in regressions involving highly serially
correlated variables. Ferson, Sarkissian, and Simin (2003) and Campbell and Yogo
(2006) show that this leads to spurious regression problems that cause the null
hypothesis to be rejected too often in the kind of regressions we are dealing with.
To address these issues, this section constructs p-values __ using a Monte Carlo pro-
cedure to find the actual distribution of the statistic ( c NcN) /cc. The aim is
T
to build a model for returns that is as close as possible to the one used in Stambaugh
(1999), but that allows for serial correlation of the error terms. To this end, we con-
sider in addition to equation (8) an equation for returns of the form
P
Rt,t+N =A N+C N___t+U Nt+N
(10) ,
Dt
with given constants A Nand C N. We allow U Nt to be serially correlated by specifying

it as an AR(1) process. Since the innovations PD t to the PD ratio in equation (8) are
likely also a key component of the innovation to returns, we allow the innovations
Nt to be correlated with PD
to U t and consider
U Nt =U Nt1
(11) +t + PD
t
for t N(0, 2 ),independent of PD

j at all dates j, and for given constants and
satisfying || < 1.
Notice that equation (10) is not a special case of equation (5) in Section IIB. The
reason is that U Nt+N is correlated with PD
t+Nj
for j0and thus correlated with PDt :
( ) N 2
________
PD
E(U Nt+N
PDt )=
.
1() 2
As a result, the regression coefficients ( a N ,c N) in equation (5) do not sat-

isfy a =A , c =C ,whenever
N N N N
0. It would thus be incorrect to plug
our estimates of a , c into equation (10) for the purpose of running the Monte
N N
Carlo simulations.
To estimate the parameters in equations (10) and (11), we proceed as follows. We
lag equation (10) by one period, multiply by and subtract it from equation (10).
This delivers
P Pt1
Rt,t+N =A N (1)+Rt1,t+N1 +C N___tC N____
(12)
t+N
+ PD +t+N ,
Dt Dt1

which can be estimated using nonlinear least squares and the observed explana-
tory variables (Rt1,t+N1 Dt Dt1 t+N )
Pt ___ P PD
,__ , t1
, , because these explanatory variables are
orthogonal to t +N. We thus have consistent and efficient estimates for , , 2 ,A N,
and C N. We plug these estimates into equations (10) and (11) to simulate R .
t,t+N
Table 1BRational Expectations Test, H 0: cc

(Stock Returns, One-Sided Test, Small Sample Corrections with Serially Correlated Errors
and Nonlinear Least Squares Estimates from Section IIC)

c 103 c 103 E(c c) H
0 : cc c 103 c 103 E(c c) H0 : cc
UBS*, >100k, 1 yr, SPF 0.82 2.80 0.64 0.054 0.014 2.68 0.62 0.062 0.013
UBS*, >100k, 1 yr, Michigan 0.82 2.79 0.64 0.054 0.016 2.67 0.62 0.071 0.014
UBS*, all, 1 yr, SPF 0.82 2.78 0.64 0.072 0.013 2.69 0.62 0.063 0.013
UBS*, all, 1 yr, Michigan 0.82 2.77 0.64 0.072 0.015 2.68 0.62 0.055 0.014
CFO, 1 yr, SPF 0.94 5.83 3.19 0.568 0.046 6.59 4.07 0.550 0.111
CFO, 1 yr, Michigan 0.94 5.90 3.19 0.568 0.044 6.67 4.07 0.520 0.107

Shiller, 1 yr, SPF 0.92 4.68 2.34 0.621 0.020 4.68 2.38 0.627 0.018
Shiller, 1 yr, Michigan 0.92 4.74 2.34 0.624 0.016 4.74 2.38 0.627 0.017
Shiller, 10 yrs, SPF 0.76 8.07 2.65 0.130 0.016 9.45 2.66 0.200 0.014
Shiller, 10 yrs, Michigan 0.76 7.48 2.65 0.200 0.023 8.85 2.66 0.200 0.019
Notes: The column labeled reports the estimated persistence parameter of the return residuals U Nt in equa
tion(11). See the notes to Table 1A for further information.
To compute expected returns under the null hypothesis of RE, we compute the
true expectation of returns, which are given by
P
Et( Rt,t+N) =A N+C N___t+() NU Nt .

Dt
Using these results, we can__simulate all the variables involved in equations (3) and
(5), compute the statistic ( c NcN)/cc
T for each simulation and study its small
sample distribution, using the sample __ sizes of the considered survey source. We then
compute the probability that T ( c NcN) /ccin the Monte Carlo simulations is
smaller than the corresponding value we find for the data. This provides a p-value
for the one-sided test of RE, when the alternative hypothesis is c N>c N, i.e., that
survey returns respond more strongly to the PD ratio than actual returns.
Table 1B reports the outcomes of this procedure. The second column in the table
reports the estimated value for . It shows that the residuals U Nt in equation (11) are
indeed serially correlated. We find that this leads to considerable __ spurious regression
problems, as the standard deviation of the test statistic T ( c NcN) /cc
is indeed
around 2 to 3 times larger than its asymptotic value of 1.
Table 1B also reports the bias corrected estimates c N and cN.31 Compared to the
results in Table 1A, the estimates for c Nare considerably less negative; the ones for
the CFO and Shiller 1 year sample even become positive. Yet, the bias-corrected
estimates for cNin Table 1B also become more positive when compared to the ones
reported in Table 1A. Therefore, despite the spurious regression problems, which
cause an increase in the true variance of the test statistic, the RE hypothesis is
31
The point estimates correct for small sample bias using the mean of the estimator found in the Monte Carlo
simulations with serially correlated errors and using the fact that c N=C N +() N (1 2 )/(1() 2 ).
Table 1CRational Expectations Test, H 0: cc

(Excess Stock Returns, One-Sided Test, Small Sample Corrections with Serially Correlated Errors
and Nonlinear Least Squares Estimates from Section IIC)

c 103 c 103 E(c c) H
0 : cc c 103 c 103 E(c c) H0 : cc
UBS, >100k, 1 yr, SPF 0.77 2.00 1.11 0.153 0.051 1.89 1.10 0.154 0.058
UBS, >100k, 1 yr, Michigan 0.77 2.00 1.11 0.157 0.056 1.89 1.10 0.158 0.054
UBS, all, 1 yr, SPF 0.77 1.98 1.11 0.161 0.056 1.90 1.10 0.154 0.057
UBS, all, 1 yr, Michigan 0.77 1.98 1.11 0.154 0.054 1.90 1.10 0.160 0.062
CFO, 1 yr, SPF 0.93 5.58 3.00 0.532 0.067 6.44 4.12 0.517 0.155
CFO, 1 yr, Michigan 0.93 5.59 3.00 0.540 0.065 6.45 4.12 0.522 0.151

Shiller, 1 yr, SPF 0.91 4.01 1.85 0.562 0.028 4.01 1.84 0.548 0.028
Shiller, 1 yr, Michigan 0.91 4.00 1.85 0.547 0.030 4.01 1.84 0.540 0.032
Shiller, 10 yrs, SPF 0.78 6.30 5.16 0.151 0.030 7.68 5.15 0.175 0.023
Shiller, 10 yrs, Michigan 0.78 6.14 5.16 0.383 0.033 7.52 5.15 0.214 0.026
Notes: The column labeled reports the estimated persistence parameter of the return residuals U Nt in equa
tion(11). See the notes to Table 1A for further information.
soundly rejected.32 The level of the rejection is now considerably lower than the one
reported in Table 1A, but still highly significant. Since the involved sample lengths
are not very large, this is a remarkable result.
Table 1C repeats the analysis when letting Rt,t+Ndenote excess stock returns rather
than stock returns.33 We construct excess return expectations following Bacchetta,
Mertens, and Van Wincoop (2009), i.e., we assume that the Nperiod ahead risk-free
interest rate is part of agents information set and subtract it from the (expected) stock
return.34 We find that the strength of the rejection of RE is then somewhat lower
when compared to Table 1B. Still, for most survey sources one obtains p-values near
or below 5 percent. The somewhat lower p-values show that the approach based on
plain stock returns, as reported in Table 1B, offers a slightly more powerful test of
the RE hypothesis. Interestingly, this is in line with the main hypothesis of this paper,
namely, that agents form their expectations about stock prices by extrapolating the
behavior of past price growth. Under this hypothesis, subtracting the risk-free interest
rate adds noise to the independent variables on the LHS of regression equations (3)
and (5), which is consistent with the observed decrease in the significance levels.
D. How Models of Learning May Help
This section illustrates that a simple adaptive approach to forecasting stock prices
is a promising alternative to explain the joint behavior of survey expectations and
stock price data.
32
As in Table 1A, the rejection is less strong for the CFO survey, especially when using the survey median. See
footnote 29 for a discussion for the potential reasons behind this result.
33
For the Shiller survey, which reports price growth expectations, Rt,t+N
now denotes excess prices growth.
34
Following Bacchetta, Mertens, and Van Wincoop (2009), we use the constant maturity interest rates available
from the FRED database at the St. Louis Federal Reserve Bank.
Figure 2 shows that the peaks and troughs of the PD ratio are located very closely
to the peaks and troughs of investors return expectations. This suggests that agents
become optimistic about future capital gains whenever they have observed high cap-
ital gains in the past. Such behavior can be captured by models where agents expec-
tations are influenced by past experience, prompting us to temporarily explore the
assumption that the log of agents subjective conditional capital gain expectations
lnEt[ Pt+1
/Pt] evolves according to the following adaptive prediction model:
/Pt] = (1g)lnEt1
(13) lnEt[Pt+1 [Pt/Pt1
] + glnPt1/Pt2
,
where g>0indicates how strongly capital gain expectations are updated in the
direction of past price growth observations. While equation (13) may appear ad hoc,
we show in Section V how a very similar equation can be derived from Bayesian
belief updating in a setting where agents estimate the persistent component of price
growth from the data. Note that equation (13) incorporates price growth observa-
tions only with a lag, in line with the theoretical model that we consider later on.
One can feed into equation (13) the historical price growth data of the S&P 500
over the postwar period. Together with an assumption about capital gain expecta-
tions at the start of the sample, this delivers a time series of implied capital gain
expectations lnEt[ Pt+1/Pt] that can be compared to the expectations from the UBS
survey.35 Setting the unobserved initial price growth expectations in 1946:I equal
to 0percent, Figure 3 reports the outcome of this procedure, when estimating the
gain parameter g using nonlinear least squares to minimize the distance between the
expectations implied by equation (13) and the observed survey expectations.36 The
resulting point estimate is given by g=0.0264and has a standard error equal to
0.00168. Figure 3 shows that the adaptive prediction model captures the behavior
of UBS expectations extremely well: the correlation between the two series is equal
to +0.91.
There also exists a strong positive relationship between the PD ratio and the
capital gains expectations implied by equation (13). Figure 4 documents this rela-
tionship for the entire postwar period by plotting the joint distribution of the capital
gains expectations (as implied by equation (13)) and the PD ratio in the data.37
When regressing the PD ratio on a constant and the expectations of the adaptive
prediction model, one obtains an R 2coefficient of 0.43; using also the square of the
expectations, the R 2rises further to 0.48.38
Interestingly, the relationship between implied price growth expectations and the
PD ratio depicted in Figure 4 seems to have shifted upward after the year 2000, as
indicated by the squared icons in the figure. Indeed running the previous regressions
on the linear and squared expectations separately before and after the year 2000, one
35
We transform the UBS survey measures of return expectations into a measure of price growth expectations
Pt+1 Dt+1 Pt+1 Dt
t+1 =___
using the identity R Pt
+___
Pt
=___
Pt
+ D__
Pt
where
Ddenotes the expected quarterly growth rate of div-
idends that we set equal to the sample average of dividend growth over 1946:I2012:I, i.e, D = 1.0048. Results
regarding implied price growth are very robust toward changing Dto alternative empirically plausible values.
36
The figure reports log quarterly expected growth rates for real stock prices.
37
As before, we set the unobserved initial price growth expectations in 1946:I equal to 0percent.
38
The p-values for the coefficients on the linear and squared expectations are all statistically significant at the
1 percent level.
0.035
Adaptive prediction model (g = 0.02641, m(1946:I) = 0%

log real price growth expectations
0.03
UBS expected price growth (own portfolio)
0.025
(per quarter)
0.02
0.015
0.01
0.005
0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Quarter
Figure 3. Real Price Growth Expectations Implied by the UBS Survey Expectations
and by the Adaptive Prediction Model (Equation (13)), 1998:II2007:IV
400
1946:I1999:IV
350 2000:I2012
300
Quarterly PD ratio
250
200
150
100
50
0.015 0.01 0.005 0 0.005 0.01 0.015 0.02 0.025 0.03
log price growth expectations
Figure 4. Quarterly PD Ratio of the S&P 500 Graphed against the Quarterly Real
Price Growth Expectations Implied by the Adaptive Prediction Model (Equation (13)),
for the Same Quarter, 1946:I2012:I
obtains much higher R 2values, namely 0.72 and 0.77, respectively. We will come
back to this issue in Section IX.
Overall, it emerges from Figure 4 that variations in expected capital gains
Et[ Pt+1
/Pt] can accountin a purely statistical sensefor a large share of the vari-
ability in the postwar PD ratio. This suggests that an asset-pricing model consistent
with equation (13), which additionally predicts a positive relationship between the
PD ratio and subjective expectations about future capital gains, has a good chance
of replicating the observed positive comovement between price growth expectations
and the PD ratio. The next sections spell out the microfoundations of such a model.
III. A Simple Asset-Pricing Model
Consider an endowment economy populated by a unit mass of infinitely lived

agents i[0,1]with time-separable preferences. Agents trade one unit of a
stock in a competitive stock market. They earn each period an exogenous nondiv-
idend income Wt 0that we refer to as wages for simplicity. Stocks deliver the
exogenous dividend Dt 0. Dividend and wage incomes take the form of perish-
able consumption goods.
The Investment Problem.Investor i solves

t u(Ct i )
0
i
(14a) E

max
{Ct i 0,St i } t=0 t=0
(Pt +Dt)+Wt for all t0,

subject to St i Pt +Cti =St i1
_
(14b) S St i S,

i1 = 1,
S
where C idenotes consumption, uthe instantaneous utility of the consumer,

assumed to be continuous, differentiable, increasing and strictly concave, S i
the agents stockholdings, _ which are subject to upper and lower limits such that
<S <1<S< , and P0 the (ex-dividend) price of the stock; i

denotes the agents subjective probability measure, which may or may not satisfy
the rational expectations hypothesis. Further details of iwill be specified below.
To simplify the exposition, we do not explicitly consider trade in risk-free bonds
(Bt i ) in problem (14), instead impose from the outset that actual bond prices and
agents subjective beliefs about future bond prices are such that agents hold and plan
to hold zero bonds at all times (Bt i 0). When bonds are in zero net supply, these
beliefs are in line with the equilibrium outcomes, when all agents are identical or
when agents face the borrowing constraint Bt i 0.39
Dividend and Wage Income.As is standard in the literature, we assume that div-
idends grow at a constant rate and that dividend growth innovations are unpredictable:
(15) lnDt = ln D + lnDt1 + ln D

t ,
where D 1denotes gross mean dividend growth and l n D

t an i.i.d. growth inno-
vation described further below.
39
In the latter case, the most patient agent prices the bond, i.e., letting R b denote the gross real return on a
i[0,1]Et u(Ct+1
one-period risk-free bond, we have 1/R b=sup )/u(Ct i ).
i
i
We also specify an exogenous wage income process Wt, which is chosen such
that the resulting aggregate consumption process C t =Wt +Dtis empirically
plausible. First, in line with Campbell and Cochrane (1999), we set the standard
deviation of consumption growth to be 1/7 of the standard deviation of dividend
growth. Second, again following these authors, we set the correlation between
consumption and dividend growth equal to 0.2. Third, we choose a wage pro-
cess in the model such that the average consumption-dividend ratio in the model
(E[Ct /Dt] )equals the average ratio of personal consumption expenditure to net divi-
dend income in US postwar data and displays persistence similar to that observed in
the data. All this can be parsimoniously achieved using the following wage income
process:
W Wt1
(16) ln(1+___t) = (1p)ln(1+)+pln(1+____
Dt1 )
+ ln W
t ,
Dt
where 1+is the average consumption-dividend ratio and p [0,1)its quarterly

persistence. The innovations are given by
(ln t ) ( 2 ( 2W) (DW 2W ))

ln D
t 1 D , D D W
2 2
W
(17) iiN __ ,

with E D
t = E t = 1. Given the variance of dividend growth
W
2D , which can be
estimated from dividend data, one can use D Wand
W to impose the desired vol-
2
atility of consumption growth and the desired correlation with dividend growth.
Choosing =22, one obtains the observed average consumption-dividend ratio in
the data and setting p to a value close to 1, one replicates the observed persistence
of the consumption-dividend ratio in the data. Online Appendix A4 provides further
details.
The Agents Underlying Probability Space.Agents hold a set of subjective

probability beliefs about all payoff-relevant variables that are beyond their con-
trol. In addition to fundamental variables (dividends and wages), agents perceive
competitive stock prices as beyond their control. Therefore, the belief system
also specifies probabilities about prices. Formally, letting denote the space of
possible realizations for infinite sequences, a typical element is given by
= {Pt,Dt,Wt } .As usual, tthen denotes the set of all possible (nonnega-
t=0
tive) price, dividend, and wage histories from period zero up to period tand t its
typical element. The underlying probability space for agents beliefs is then given
by (, , i )with denoting the corresponding -Algebra of Borel subsets of ,
ia probability measure over (,).
and
The agents plans will be contingent on the history t, i.e., the agent chooses
state-contingent consumption and stockholding functions
(18) Cti : t +,
_
St i : t[S, S].
(19)

The fact that C iand S idepend on price realizations is a consequence of optimal

choice under uncertainty, given that agents consider prices to be exogenous random
variables.
The previous setup is general enough to accommodate situations where agents
learn about the stochastic processes governing the evolution of prices, dividends,
and wages. For example, imay arise from a stochastic process describing the evo-
lution of these variables that contains unknown parameters or hidden variables about
which agents hold prior beliefs. The presence of unknown parameters or hidden
variables then implies that agents update their beliefs using the observed realizations
of prices, dividends, and wages. A particular example of this kind will be presented
in Section V when we discuss learning about stock price behavior.
The probability space defined above is more general than that specified in a
RE analysis of the model, where contains usually only the variables that are
exogenous to the model (in this case D tand Wt), but not variables that are endoge-
nous to the model and exogenous to the agent only (in this case Pt). In the RE lit-
erature, the Bayesian RE literature, or the literature modeling robustness concerns,
agents are assumed to know the equilibrium pricing function Pt ((D,W) t ) mapping
histories of dividends and wages into a market price. Prices then carry only redun-
dant information and can be excluded from the probability space without loss of
generality. The more general formulation we entertain here allows us to consider
agents who do not know exactly which price materializes given a particular history
of dividends and wages; our agents do have a view about the distribution of P t con-
ditional on (D,W) ,but in their minds this is a proper distribution, not a point mass
t
as in the RE case. Much akin to academic economists, investors in our model have
not converged on a single asset-pricing model that associates one market price with
a given history of exogenous fundamentals.
Parametric Utility Function.To obtain closed-form solutions, we consider in

the remaining part of the paper the utility function
Ct 1

u(Ct )=____
(20)
with >1.
1
We furthermore assume that
RE < 1,
(21)
t ) ( t )
1
RE =( D ) 1 E[( W
where D
]. This insures existence of an equilibrium
under rational price expectations.
Existence of a Recursive Solution.Since solving the optimization problem (14)

for general (potentially nonrational) price beliefs is nonstandard, it is worth pointing
out that this problem is well defined. Existence of a maximum is guaranteed by the
stock limits (14b), which ensure that the choice of stocks is made over a compact
set, in combination with a bounded continuous objective function (20), satisfying
u(Ct )0. Sufficiency of first-order conditions (FOC) is guaranteed because the
agents problem is concave. Given the subjective price beliefs introduced in the
remaining part of the paper, agents posterior will follow a recursive form. Using a
standard normalization of the utility function and of the budget constraint, one can
then guarantee that the optimal solution to (14) takes the form
P W
St i =S i(St1
(22) i,___t,___t,mti ),
Dt Dt
ti is a sufficient statistic characterizing the subjective distributions about
where m
future values of (____ D D )
Pt+j Wt+j

,____ ,___for j>0.40 See online Appendix A5 for details.
Dt+j

D
t+j1 t+j t+j
IV. Rational Expectations Equilibrium
As a point of reference, we determine the stock price under the assumption of RE.
For the limiting cases with an i.i.d. consumption-dividend ratio (p=0) or with a
unit root in this ratio (p1), one can derive simple closed-form expressions for
the RE stock price: see online Appendix A6.
Proposition 2: If agents hold rational expectations and if price expectations

satisfy the usual transversality condition (stated explicitly in online Appendix A6),
then the RE equilibrium price dividend ratio is

Pt RE RE
(23) ____ =_______
for p1,
Dt 1 RE
Pt RE
RE ( t )
RE
(24) ____ = b_______
W
for p=0,
Dt 1
t ) ( t ) t ) ( t )
1 1 D
__
2
where RE ( D ) 1 E[( W D
] , bE[( W D
]e (1) 2, and
( D ) 1e (1) D/2.
RE 2
The previous result shows that under RE the equilibrium PD ratio inherits the
persistence properties of the consumption-dividend ratio process. Specifically, for
p=0 , the PD ratio is an i.i.d. process, thus, fails to match the persistence of the PD
ratio observed in the data. For the empirically more plausible case with a persistent
consumption-dividend ratio (pclose to 1), the PD ratio also becomes persistent but
volatility then strongly falls. In the limit p1, which is the case considered in
Campbell and Cochrane (1999), the RE PD ratio becomes a constant. Price growth
is then given by
+1 lnPt = ln + ln t+1
(25) lnPt RE RE D D
,
40
The solution for optimal consumption plans follows from (22) and the flow budget constraint.
so that one-step-ahead price growth and return expectations are constant over time.
Moreover, as long as p is bounded below 1 , the RE equilibrium implies a nega-
tive correlation between the PD ratio and expected future returns, contrary to what
is evidenced by survey data.41 The RE equilibrium thus has difficulties in jointly
matching the persistence and volatility of the PD ratio and the behavior of survey
returns.
V. Learning about Capital Gains and Internal Rationality
This section specifies agents beliefs. Once one departs from rational expectations,
these beliefs become part of the microfoundations of the model. The benchmark
beliefs presented in this section aim at specifying price beliefs in a parsimonious
way such that they have a chance of replicating the behavior of survey expectations.
We study alternative and more elaborate belief formulations in Sections XB and XC.
A. Agents Belief System: Benchmark Specification
To focus on the effects of learning about price behavior, we assume that agents
know the processes (15)(17), i.e., agents hold rational dividend and wage expecta-
tions. This sets us apart from the Bayesian RE literature, which focuses on learning
about the behavior of exogenous fundamentals.
We also have to specify price beliefs. Our specification is motivated by the fol-
lowing observations. Under RE, price growth is approximately equal to dividend
growth, whenever the consumption-dividend ratio is persistent (pclose to 1): see
Proposition 2. Price growth in the data, however, can persistently outstrip dividend
growth, thereby giving rise to a persistent increase in the PD ratio and an asset price
boom (see Figure 1); conversely, it can fall persistently short of dividend growth
and give rise to a price bust. This behavior of actual asset prices suggests that it is of
interest to relax agents beliefs about price growth behavior. Indeed, in view of the
actual behavior of asset prices, agents may entertain a more general model of price
behavior, incorporating the possibility that the growth rate of prices persistently
exceeds/falls short of the growth rate of dividends. To the extent that the equilib-
rium asset prices implied by these beliefs display such data-like behavior, agents
beliefs will be generically validated.
In line with the previous discussion, our benchmark assumption is that agents
perceive prices to evolve according to
(26) lnPt+1 lnPt = lnt+1

+lnt+1
,
where t +1denotes a transitory shock to price growth and a persistent price

t+1
growth component that drifts slowly over time according to
(27) lnt+1 = lnt + lnt+1 .
Online Appendix A6 derives the general expression for the RE price for the case p(0,1).
41
For simplicity, we assume that agents perceive the innovations lnt and lnvtto be
jointly normally distributed according to
__

2
2 2 0
( 0 2)
lnt
(lnt)
(28)
iiN 2 , .
__

2

Since agents observe stock price growth, but do not separately observe the persistent
and transitory subcomponents driving it, the previous setup defines a filtering prob-
lem in which agents need to decompose observed price growth into its persistent and
transitory elements, so as to forecast optimally.42
The presented belief setup has a number of appealing features. First, it allows
capturing the fact that stock price growth displays large transitory fluctuations by
imposing 2 << 2. In the limiting case where 2is close to 0, the persistent
price growth component behaves almost like a constant, as in the RE solution with
p1.The subjective price beliefs specified above can thus be interpreted as a
small deviation from RE equilibrium beliefs. Second, the benchmark setup can cap-
ture periods with sustained increases in the PD ratio (t+1 > D )and sustained
decreases (t +1 < D ), in line with the behavior of the PD ratio in the data. To the
extent that the PD ratio in the model also displays such behavior, agents beliefs
about the presence of a persistent price growth component will also be generically
validated by the model outcome. Third, as we shall show in the next section, the
belief specification implies that agents perceived price growth expectations will
approximately evolve according to equation (13), which captures the time series
behavior of survey expectations.
As with any assumption about model primitives, one can entertain other plausi-
ble alternatives. Section XB explores alternative ways to specify the belief system,
which incorporate mean reversion in the PD ratio.
B. Efficiency of Stock Prices and Internal Rationality
Among academics there appears to exist a widespread belief that rational behav-
ior and knowledge of the fundamental processes (dividends and wages in our case)
jointly dictate a certain process for stock prices and thus the price beliefs that agents
can rationally entertain.43 This view stipulates that rational behavior implies knowl-
edge that current stock prices must equal a discounted sum of dividends. Individual
rationality and rational expectations about fundamentals would then provide inves-
tors with knowledge of the equilibrium pricing function (as is assumed under RE or
Bayesian RE), so that postulating subjective price beliefs, e.g., those specified in the
previous section, would be inconsistent with the assumption of optimal behavior on
the part of agents.
42
Note that we do not incorporate mean-reversion into price growth beliefs in our benchmark setting. This is for
simplicity, as we wish to consider the most parsimonious way to include perceived booms and busts. In addition, we
seek to determine the model-endogenous forces that lead to a reversal of boom and bust dynamics, i.e., we do not
want to obtain reversals because they are hard-wired into beliefs. We extend the setup to one with mean reversion
in Section XB.
43
We often received this reaction during seminar presentations.
This view is correct only in special cases. Considering the case with risk-neutral
agents, Adam and Marcet (2011) show that it is correct only if agents are homoge-
neous and do not face trading constraints. In a setting with heterogeneous r isk-neutral
agents and trading constraints, it fails to be correct. As we show below, agents in our
model are internally rational even in the homogeneous agent case: their behavior is
optimal given an internally consistent system of subjective beliefs about variables
that are beyond their control, which includes prices. All that is needed is a concave
utility function.44
To illustrate this point, consider first risk-neutral agents with rational dividend
expectations and ignore limits to stock holdings. Forward-iteration on the agents
optimality condition
Pt+1 +Dt+1
u(Ct i )=Et [u(Ct i+1
(29) )________
]

i
P t
and risk-neutrality deliver the present value relationship
[i=1 ]
T
Pt =Et
iDt+i + TEt [Pt+T],
i

which is independent of the agents own choices. Provided agents price beliefs
satisfy a standard transversality condition (lim
T
T i
Et [
P ]
t+T = 0for all i ), each

rational agent would conclude that there must be a degenerate joint distribution for
prices and dividends given by

[i=1 ]
Pt =Et
(30) iDt+i a.s.
Since the RHS of the previous equation is fully determined by dividend expecta-
tions, the beliefs about the dividend process deliver the price process compatible
with optimal behavior. In such a setting, it would be plainly inconsistent with opti-
mal behavior to assume the subjective price beliefs (26)(27).45
Next, consider a concave utility function u()satisfying standard Inada condi-
tions. Forward iteration on (29) and assuming an appropriate transversality condi-
tion then delivers

[ j=1 ]
Pt u(Ct )=Et
(31)
j u(Ct+j
) Dt+j
a.s.
Unlike in equation (30), the right side of the previous equation depends on the
agents subjective consumption plans. From equation (18) follows that future con-
sumption plans are of the form Ct+j ((P,D,W ) t+j ).Expected future consumption
44
Obviously, the subsequent discussion takes for granted the fact that investors are homogeneous is not com-
mon knowledge.
45
See Adam and Marcet (2011) for a discussion of how in the presence of trading constraints, this conclusion
breaks down with linear consumption preferences.
thus depends on the agents price expectations, say those implied by equation (26).
Indeed, whatever are the agents price expectations, subjective consumption plans
will adjust, so as to satisfy the first-order condition (29). As a result, equation (31)
will also hold. Equation (31) thus fails to impose any restriction on what optimizing
agents can possibly believe about the price process, given their knowledge about the
tand D
processes for W t. With the considered nonlinear utility function, we can thus
simultaneously assume that agents maximize utility, hold the subjective price beliefs
(26)(27) and have rational expectations about dividends and wages.
The fact that equilibrium prices can be written as a discounted sum of the form
(31) differs notably from the standard pricing formula studied in modern asset-
pricing theory, which determines equilibrium prices using

[ j=1 ]
Pt u(Wt +Dt )=Et

j u(Wt+j +Dt+j ) Dt+j a.s.
The RHS of this equation uses equilibrium values of future consumption instead of
the subjective future consumption plans showing up in equation (31). In this sense,
stock prices satisfying (31) are not efficient, because they do not discount dividends
with the equilibrium discount factor.
C. Learning about Capital Gains
Under internal rationality, the specification of the perceived price process (26)
dictates the way that agents learn from observed prices. The presence of an unob-
served permanent component in equation (26) gives rise to an optimal filtering prob-
lem. To obtain a parsimonious description of this problem, we specify conjugate
prior beliefs about the unobserved persistent component l nt at t =0. Specifically,
agent is prior is
(32) ln0 N(lnm0i , 2 ),
where prior uncertainty 2is assumed to be equal to its Kalman filter steady-state
value, i.e.,
_________
2
(33) . _______________
( )
2+
2 + 4 2 2
2
Equations (26), (27), and (32), and knowledge of the dividend and wage income
i.
processes (17) then jointly specify agents probability beliefs
The optimal Bayesian filter then implies that the posterior beliefs following some
tare given by46
history
(34) lnt | t N(lnmti , 2 ),
See West and Harrison (1997, Theorem 3.1). Choosing a value for 2different from the steady-state
46
value (33) would only add a deterministically evolving variance component 2t to posterior beliefs with the
property lim
tt = , i.e., it would converge to the steady-state value.
2 2
with
___v+ g(lnPt lnPt1 +______ )

2 2 + 2v
lnmti =lnmt1
(35) i
lnmt1
i

,
2 2
(36) g= ______ 2
.
+ 2
2
Agents beliefs can thus be parsimoniously summarized by a single state variable

(mti ) describing agents degree of optimism about future capital gains. These beliefs
evolve recursively according to equation (35) and imply that
i P
(37) ln mti = lnEt [____
Pt ]
t+1
2/2,
so that equation (35) isup to some (small) variance correction and the pres-
ence of a time lagidentical to the adaptive prediction model (13) considered in
SectionIID.
The subjective price beliefs (26), (27), and (32) generate perfect foresight equi-
librium price expectations in the special case in which prior beliefs are centered at
the growth rate of dividends, i.e.,
lnm0i = ln D ,
and when considering the limiting case with vanishing uncertainty, where
( 2 , 2 , 2D , 2W)
0. Agents prior beliefs at t=0about price growth in t1
then increasingly concentrate at the perfect foresight outcome ln D (see equations
(26) and (27)). With price and dividend expectations being at their perfect fore-
sight value, the perfect foresight equilibrium prices implied by Proposition 2 for
the vanishing noise limit become the equilibrium outcome at t=0. Importantly,
it continues to be possible to study learning dynamics in the limit with vanishing
risk: keeping the limiting ratio 2 / 2finite and bounded from zero as uncertainty
vanishes, the Kalman gain parameter g,defined in equation (36), remains well spec-
2 g 2
ified and satisfies lim__2 = lim___
. We will exploit this fact in Section VII when
1g
presenting analytical results.
VI. Dynamics under Learning
This section explains how equilibrium prices are determined under the subjective
beliefs introduced in the previous section and how they evolve over time.
Agents stock demand is given by equation (22). Stock demand depends on the
belief mti , which characterizes agents capital gains expectations. These beliefs
evolve according to (35). As a benchmark, we shall now assume that all agents
hold identical beliefs (mti =mtfor all i). In doing so, we assume that the marginal
investor has return expectations in line with those documented for the survey data.
We consider the homogeneous agent case first, so as to stay as close as possible to
the representative agent settings typically considered under RE. Section VIIIB con-
siders extensions to settings with heterogeneous beliefs.
Using the representative agent assumption, the fact that stocks are in unit supply
and imposing market clearing in periods tand t1in equation (22), one obtains
that the equilibrium price in any period t0 solves
P W
(38) 1=S(1,___t,___t,mt).
Dt Dt
The beliefs mtand the price dividend ratio Pt /Dtare now simultaneously deter-
mined via equations (35) and (38). Due to a complementarity between realized
capital gains and expected future capital gains,47 this simultaneity can give rise to
multiple market-clearing price and belief pairs. In the learning literature, the stan-
dard approach to resolve this issue consists of assuming that agents use only lagged
endogenous variables to update model estimates (Eusepi and Preston 2011; Evans
and Honkapohja 2001; Marcet and Nicolini 2003; Sargent, Williams, and Zha 2009).
In our benchmark specification we shall follow this approach, but we provide below
Bayesian foundations for lagged updating in a setting where agents can observe
current prices. Furthermore, as we show in Section XC, lagged belief updating is
objectively optimal because agents who update beliefs using current price informa-
tion experience lower utility in equilibrium.
Introducing an updating lag into equation (35) can be justified as internally ratio-
nal by slightly modifying the information structure. The modification is relatively
straightforward and consists of assuming that agents observe at any time t informa-
tion about the lagged transitory price growth component t1entering equation (26).
In such a setting, it is optimal to give less weight to the last available price growth
observation, when estimating the permanent component t. This is so because the
last observation is more noisy. Formally, online Appendix A7 shows that in the limit
where agents learn all transitory factors with a lag, Bayesian updating implies
(39) lnmt = lnmt1 + g(lnPt1 lnPt2 lnmt1) + gln 1t ,
where updating now occurs using only lagged price growth (even though agents do
observe current prices) and where ln 1t iiN(___ , 2 )is a time tinnovation to
2
2

agents information set (unpredictable using information available to agents up to
period t1). The shock ln 1t thereby captures the information that agents receive
in period tabout the transitory price growth component lnt 1.
With this slight modification, agents beliefs mtare now predetermined at time t ,
so that the economy evolves according to a uniquely determined recursive process:
equation (38) determines the market-clearing price for period tgiven the beliefs mt
and equation (39) determines how time tbeliefs are updated following the observa-
tion of the new market-clearing price.48
47
Intuitively, a higher PD ratio implies higher realized capital gains and thus higher expectations of future gains
via equation (35). Higher expected future gains may in turn induce a higher willingness to pay for the asset, thereby
justifying the higher initial PD ratio.
48
There could still be an indeterminacy arising from the fact that S ()is nonlinear, so that equation (38) may
not have a unique solution. We have not encountered such problems in our analytical solution or when numerically
solving the model.
VII. Equilibrium: Analytic Findings
This section derives a closed-form solution for the equilibrium asset price for
the special case where all agents hold the same subjective beliefs and where
these beliefs imply no (or vanishing) uncertainty about future prices, dividends,
and wages. While the absence of uncertainty is unrealistic from an empirical stand-
point, it helps us in deriving key insights into how the equilibrium price depends on
agents beliefs, as well as on how prices and beliefs evolve over time.49 The empir-
ically more relevant case with uncertainty will be considered in Section IX using
numerical solutions.
We present a series of results that increasingly adds assumptions on agents beliefs
system . The next section provides a closed-form expression for the equilibrium
PD ratio as a function of agents subjective expectations about future stock market
returns for any belief system without uncertainty. Section VIIB then discusses the
pricing implications of this result for the subjective capital gains beliefs presented
in Section V. Finally, Section VIIC shows how the interaction between asset price
behavior and subjective belief revisions can temporarily delink asset prices from
their fundamental value, i.e., give rise to a self-reinforcing boom and bust cycle in
asset prices along which subjective expected returns rise and fall.
A. Main Result
t+1 (Pt+1 +Dt+1)/Ptdenote the stock return, we have following

Letting R
main result.50
Proposition 3: Suppose u(C)=C 1 /(1), agents beliefs imply no

uncertainty about future prices, dividends, and wages. Assume given posterior
beliefs about prices, dividends and wages that satisfy
( j=1(i=1 Rt+i ) t+j)

T j
(40)
lim t RT > 1 and lim
E
t
E
____
1W <,
T T
_
then, if S ,| S| are sufficiently large, the equilibrium PD ratio in period t is given by
| |

1_1
( ( t R ) )

P W
(1+___t)
( ) E
j j
(41) ___t= ____
_ 1
1
D D t t j=1 i=1 t+i
( j=1(i=1 Rt+i ) )
j
1E
___ ____
1W .
Dt t t+j
49
An analytic solution can be found because in the absence of uncertainty one can evaluate more easily the
expectations of nonlinear functions of future variables showing up in agents FOCs.
50
The proof can be found in online Appendix A8.
Conditions (40) insure that the infinite sums in the pricing equation (41)
c onverge.51 Under the additional assumption that agents hold rational wage and
dividend expectations and that Wt /Dt = , equation (41) simplifies further to52
1
___

(( ) Et ____ )
( i=1 Rt+i )

j j
(1+)
P
(42) ___t=
_1 1
Dt j=1
( j=1 ( t i=1 Rt+i ))

( D) E
j
____
j
1 .
We now discuss the implications of equation (42), focusing on the empirically rele-
vant case where >0and >1.
Consider first the upper term on the RHS of equation (42), which is decreasing
in the expected asset returns. This emerges because for >1the wealth effect
of a change in return expectations then dominates the substitution effect, so that
expected asset demand and therefore the asset price has a tendency to decrease as
return expectations increase. The negative wealth effect thereby increases in strength
if the ratio of wage to dividend income () increases. This is the case because higher
return expectations also reduce the present value of wage income.
Next, consider the lower term on the RHS of equation (42), including the neg-
ative sign premultiplying it. This term depends positively on the expected returns
and captures a substitution effect that is associated with increased return expec-
tations. This substitution effect only exists if >0, i.e., only in the presence of
nondividend income, and it is increasing in . It implies that increased return expec-
tations are associated with increased stock demand and thus with a higher PD ratio
in equilibrium. It is this term that allows the model to match the positive correlation
between expected returns and the PD ratio.
This substitution effect is present even in the limiting case with log consumption
utility (1). The upper term on the RHS of equation (42) then vanishes because
the substitution and wealth effects associated with changes in expected returns can-
cel each other, but the lower term still induces a positive relationship between prices
and return expectations. The substitution effect is also present for >1and can
then dominate the negative wealth effect arising from the upper term on the RHS of
(42). Consider, for example, the opposite limit with . Equation (42) then
delivers
) ( i=1 Rt+i )

j=1(
P
1( D) ) Et ____
(
j
t=
___ 1+
j
1.
Dt j=1
Since
D > 1, there is a positive relationship between prices and expected asset
returns, whenever is sufficiently large. The two limiting results (1 and
51
These are satisfied, for example, for the expectations associated with the perfect foresight RE solution.
Equation (41) then implies for p =1that the PD ratio equals the perfect foresight PD ratio, i.e., the value given
by equation (24) with W
t = 1, as is easily verified. Conditions (40) are equally satisfied for the subjective beliefs
defined in Section V, when considering the case with vanishing uncertainty ( 2 , 2 , 2D , 2W )0.
In deriving equation (42) we abstract from transitional dynamics in Wt/Dtand set W
52
t/Dt = .
450
Perfect foresight
Capital gain (D 1)
400
350
Quarterly PD ratio
300
250
200
150
Perfect foresight PD ratio
100
50
0.03 0.02 0.01 0 0.01 0.02 0.03 0.04 0.05
Expected capital gains (mt 1)
Figure 5. Quarterly PD Ratio and Expected Capital Gains

(Vanishing Noise Limit, Estimated Model from Table 3, Diagonal Matrix)
) thus suggest that for sufficiently large the model can generate a positive
relationship between return expectations and the PD ratio, in line with the evidence
obtained from survey data.
B. PD Ratio and Expected Capital Gains
Since future stock returns depend on the belief system and on current PD
ratio, equation (42) does not yet give an explicit solution for the PD ratio.
For the belief system t [Pt+i] =(mt) iPt
introduced in Section V,53 we have E
and Et Dt+i =( ) Dt, so that without uncertainty
D i
1
Et Pt+i1 i1 D Dt
(mt +( /mt) Pt ) .
Et Rt 1 ____________ ___
+i =
= D

Et Pt+i +Et Dt+i
Substituting this into (42) gives a nonlinear relationship between the PD ratio and
the subjective capital gain expectations m t. Since a closed-form solution for the PD
ratio is unavailable, we show numerical solutions of this equation.
Figure 5 depicts the relationship between the PD ratio and mtusing the param-
eterization employed in our quantitative application later on, but abstracting from
future uncertainty.54 Figure 5 shows that there is a range of price growth beliefs
around the perfect foresight value (mt = D ) over which the PD ratio depends
positively on expected price growth, similar to the positive relationship between
expected returns and the PD ratio derived analytically in the previous section. Over
this range, the substitution effect dominates the wealth effect because our calibra-
tion implies that dividend income finances only a small share of total consumption
(approximately 4.3 percent). As a result, stock market wealth is only a small share
Online Appendix A9 proves that condition (40) is satisfied for all beliefs mt > 0.
53
54
The parameter values are given by those listed in Table 2 and by the estimated parameters for the model with
diagonal matrix from Table 3, i.e., g=0.262 , =2.03 , =0.99514 , p=0.95.
Table 2Model Calibration
Parameter Value Calibration target

D
1.0048 Average quarterly real dividend growth
D 0.0192 Standard deviation quarterly real dividend growth
22 Average consumption-dividend ratio
DW 3.51 10 4 Jointly chosen such that corrt(Ct+1/Ct,Dt+1 /Dt)=0.2
W 0.0186 and stdt(Ct+1
/Ct)=(1/7)stdt(Dt+1
/Dt)
0.0816 SD of quarterly real stock price growth
of the total present value of household wealth (the same 4.3 percent) when beliefs
assume their perfect foresight value (mt = D).
Figure 5 also reveals that there exists a capital gains belief beyond which the PD
ratio starts to decrease for higher m. Mathematically, this occurs because if mt ,
Et i=1___ 0.
j 1
expected returns also increase without bound,55 so that Rt+i
__
Pt
From equation (42) one then obtains D 0.
t
The economic intuition for the existence of a maximum PD ratio is as follows: for
tthe present value of wage income is declining, as increased price growth
higher m
optimism implies higher expected returns56 and therefore a lower discount factor.
This can be seen by noting that the FOC (29) can alternatively be written as

1=Et [( Rt+1],
Ct+1
Ct )
____
which implies that increased return expectations E t Rt+1imply a lower discount

factor Et [(Ct+1 /Ct) ]. With increased optimism, the present value of wage
57
income thus falls. At the same time, stock market wealth initially increases strongly.
Indeed, at the maximum PD ratio, stock market wealth amounts to approximately
4.5 times the value it assumes in the perfect foresight solution (see Figure 5). This
relative wealth shift has the same effect as a decrease in the wage to nonwage income
ratio
. As argued in Section VIIA, for sufficiently small values of the income
effect starts to dominate the substitution effect, so that prices start to react negatively
to increased return optimism.
C. Endogenous Boom and Bust Dynamics
We now explain how the interplay between price realizations and belief updat-
ing can temporarily delink asset prices from their fundamental values. This process
emerges endogenously and takes the form of a sustained asset price boom along
which expected returns rise and that ultimately results in a price bust along which
expected returns fall. This feature allows the model to generate volatile asset prices
and to capture the positive correlation between expected returns and the PD ratio.
P +D P
55
This follows from Et Rt+i+1 =Et _________

t+i+1
P
>
t+i+1
Et ____
t+i+1
P
=

mt.
t+i t+i
56
This is shown in online Appendix A11, which depicts the relationship between expected capital gains and
expected returns at various forecast horizons.
57
This holds true under the maintained assumption of no or vanishing uncertainty.
Consider Figure 5 and a situation in which agents become optimistic, in the sense
that their capital gains expectations mtincrease slightly above the perfect foresight
t1 = Dentertained in the previous period.58 Figure 5 shows that this
value m
increase in expectations leads to an increase in the PD ratio, i.e., Pt/Dt >Pt1/Dt1
.
Moreover, due to the relatively steep slope of the PD function, realized capital gains
will strongly exceed the initial increase in expected capital gains. The belief updat-
ing equation (39) then implies further upward revisions in price growth expectations
and thus further capital gains, leading to a sustained asset price boom in which the
PD ratio and return expectations jointly move upward.
The price boom comes to an end when expected price growth reaches a level
close to where the PD function in Figure 5 reaches its maximum.59 At this point,
stock prices grow at most at the rate of dividends ( D ), but agents hold consider-
ably more optimistic expectations about future capital gains (mt > D). Investors
high expectations will thus be disappointed, which subsequently leads to a reversal.
During a price boom, expected price growth and actual price growth thus mutually
reinforce each other. The presence of an upper bound in prices implies, however,
that the boom must come to an end. Price growth must thus eventually become very
low, sending actual and expected stock price growth eventually down.
The previous dynamics are also present in the stochastic model considered in the
next sections. They introduce low frequency movements in the P Dratio, allowing
the model to replicate boom and bust dynamics and thereby empirically plausible
amounts of asset price volatility, despite assuming standard consumption prefer-
ences. These dynamics also generate a positive correlation between the PD ratio and
expected returns.60
In the deterministic model, however, the dynamics for the PD ratio tend to be
temporary phenomena because beliefs tend to converge to the perfect foresight
equilibrium.61
Lemma 1: Consider the limiting case without uncertainty and suppose investors
hold rational dividend and wage beliefs:
(i) For any mt > 0 , we have lim m

t t = , whenever lim m
D
t t exists.

t sufficiently close to D and g<

(ii) For m _ , we have lim m
t t = if
1 D
2
(43)
D _______
_______
1<
PD( )
D
PD(m)

< 1,
m m= D |
58
In the model with uncertainty, such upward revisions can be triggered by fundamentals, e.g., by an exception-
ally high dividend growth realization in the previous period, which is associated with an exceptionally high price
growth realization.
59
In the model with noise, fundamental shocks, e.g., a low dividend growth realization, can cause the process
to end well before reaching this point.
60
While the arguments above only show that expected capital gains correlate positively with the PD ratio,
online Appendix A11 shows that expected capital gains and expected returns comove positively, so that expected
returns also comove positively with the PD ratio.
61
The proof of Lemma 1 can be found in online Appendix A10.
Table 3Asset-Pricing Moments, Data, and Estimated Models
US data Subjective belief

1946:I2012:I Subjective belief model model
(quarterly real values) (efficient matrix) (diagonal matrix)
Moment Moment t-ratio Moment t-ratio
E[PD] 139.8 120.1 0.79 115.2 0.98

std[PD] 65.2 95.6 2.06 88.2 1.56
corr[PDt , PDt1] 0.98 0.98 0.65 0.98 0.90
std[R] 8.00 8.88 2.20 7.74 0.64
c 0.0041 0.0055 1.11 0.0050 0.77
R2 0.25 0.23 0.13 0.20 0.34
E[R] 1 1.89 1.91 0.04 1.82 0.15
E[R b] 1 0.13 1.01 5.16 0.99 5.08
UBS survey data

corr[PDt,Et Rt,t+4] 0.79 0.80 0.07 0.79 0.01
Estimates
g 0.0282 0.0262
0.99524 0.99514
2.05 2.03
p 0.95 0.95
Notes: The table reports US asset-pricing moments (column 2) using the data sources described in online
AppendixA1, the moments of the estimated models (columns 3 and 5) and t-ratios for the model moments (col-
umns 4 and 6) from the SMM estimation. t-ratios are defined as (data momentmodel moment)/(estimated stan-
dard deviation of the model moment). The reported moments are as follows: E[PD], std(PD) and corr[PDt, PDt1]
denote the mean, standard deviation, and autocorrelation of the quarterly price dividend ratio, respectively; E[R]
and std[R] denote the mean and standard deviation of the gross real quarterly stock return, expressed in percentage
points, respectively; E[R b] is the gross mean risk-free interest rate, expressed in percentage points; c and R2 denote,
respectively, the regression coefficient and R2 value obtained in equation (5) when Rt,t+N is excess stock returns and
N=5 years; corr[PDt,Et Rt,t+4] denotes the correlation between the PD ratio and the subjective return expecta-
tions, with these being measured in the data column using the mean of the UBS survey (own portfolio), deflated by
the mean inflation expectations from the SPF survey. The estimated parameters are the updating gain g from equa-
tion (39), the time discount factor , the coefficient of relative risk aversion and the persistence p of the wage-div-
idend ratio from equation (16).

where P D(m) is the equilibrium PD ratio associated with beliefs m , as
implied by equation (42).
The first result in the lemma provides a global convergence result. It shows that if
beliefs settle down in this economy, they must settle down on the perfect foresight
equilibrium value. When this is the case, equilibrium prices also converge to the
perfect foresight value. While technically one cannot rule out convergence to deter-
ministic or chaotic cycles, the second result in the lemma shows that locally beliefs
do converge to the perfect foresight equilibrium, whenever the elasticity of the PD
ratio with respect to price growth believes is below 1 in absolute value and the gain
parameter g not too large. Condition (43) is satisfied, for example, for the parame-
terization of the estimated models reported in Table 3.
To illustrate the global belief dynamics further, Figure 6 depicts how beliefs
evolve over time using the parameterization of the estimated model from Table 3.62
The arrows in the figure indicate, starting from any point (mt ,mt1
) in the plane,
The vector field uses the parameters from the estimated model with diagonal matrix and estimated p reported
62
in Table 3. The vector fields for the other models from Table 3 look almost identical.
1.04
1.03
1.02
mt1
1.01
0.99
0.98
0.98 0.99 1 1.01 1.02 1.03 1.04
mt
Figure 6. Illustration of Global Belief Dynamics under Learning in the Absence of

Shocks (Estimated Model from Table 3, Diagonal Matrix)
Note: The black dot indicates the perfect foresight equilibrium belief.
the direction in which the belief pair (mt ,mt1) evolves.63 The black dot indicates
the position of the perfect foresight equilibrium (mt =mt1 = D ), which is a
rest point of the dynamics. In line with what we find when simulating the model,
Figure6 strongly suggests that beliefs globally converge to the perfect foresight
equilibrium in the absence of stochastic disturbances.
Figure 6 also shows that it takes time for beliefs to settle down at the perfect
foresight equilibrium and that agents will make persistent forecast errors along the
transition path. In particular, when agents are optimistic and stock are prices high
(mt > D ), there is a tendency for prices and beliefs to return toward their perfect
foresight values, so that realized capital gains are low. Similarly, when agents are
pessimistic and stock prices low (mt < D ), the tendency of prices and beliefs to
return to the perfect foresight values implies that realized capital gains are high.
Note that this pattern of forecast errors is consistent with the one documented for
the survey data in Section II.
VIII. Matching Asset-Pricing Moments
This section evaluates the ability of the model to replicate key asset-pricing
moments when using dividend and wage shocks as fundamental driving forces.
63
To increase readability of the graph, the length of the arrows lis nonlinearly rescaled by dividing by l 3/4 and
then linearly adjusted, so as to fit into the picture.
The set of data moments that we seek to replicate is listed in the second column
of Table 3. The first eight asset-pricing moments listed in the table are standardly
used in the asset-pricing literature to summarize the main features of stock price
volatility.64 They include the mean, standard deviation, and autocorrelation of the
quarterly PD ratio (E[PD], std[PD], and corr[PDt , PDt1 ], respectively), the mean
and standard deviation of quarterly real stock returns (E[r s ] and std[r s ]), the risk-
free interest rate (E[r b ]) and the regression coefficient (c) and R-squared value (R 2)
obtained in equation (5) when Rt,t+Nis the five year ahead excess return.65 As is
well known, it is difficult for RE models with time separable utility functions to
jointly match these moments.
We augment the standard set of moments in Table 3 by the correlation
between the PD ratio and expected stock returns from the UBS survey, denoted
corr[PDt ,E t Rt+1 ], so as to capture the behavior of survey data.
Numerically solving the nonlinear asset-pricing model with subjective beliefs
turns out to be computationally time-consuming, despite the fact that we extensively
rely on parallelization in the solution algorithms.66 For this reason, we match the
model to the data by calibrating most parameters and estimate only a key subset
using the simulated method of moments (SMM).
Table 2 reports the calibrated parameters and the calibration targets.67 The mean
and standard deviation of dividend growth ( Dand D) are chosen to match the cor-
responding empirical moments of the US dividend process. The ratio of nondividend
to dividend income () is chosen to match the average dividend-consumption
ratio in the United States for 19462011.68 The standard deviation of wage inno-
vations (W ) and the covariance between wage and dividend innovations (D W)
are chosen, in line with Campbell and Cochrane (1999), such that the correlation
between consumption and dividend growth is 0.2 and the standard deviation of con-
sumption growth is one-seventh of the standard deviation of dividend growth.69
The perceived uncertainty in stock price growth () is set equal to the empirical
standard deviation of stock price growth.70
This leaves us with four remaining parameters: the updating parameter g , the
time discount factor , the risk aversion parameter , and the persistence parameter
for the wage-dividend ratio p. Letting =(g,,,p)denote this set of parameters
and the set of admissible values, the SMM estimate is given by
(44) [()

=argmin ]
[ )],
(

64
These moments are also considered in Adam, Marcet, and Nicolini (2016).
65
The regression also contains a constant whose value is statistically insignificant and not reported in the table.
66
The numerical solution is obtained by numerically determining the stock demand function (22) solving the
FOC (29) under the subjectively perceived dividend, wage, and price dynamics, where agents understand that their
beliefs evolve according to (39). We verify that in the limiting case without uncertainty, our numerical solution
algorithm recovers the analytical solution derived in Proposition 3. Furthermore, in the case with uncertainty, we
insure the accuracy of the numerical solution by verifying that the Euler equation errors are in the order of 1 0 5 over
the relevant area of the state space. Insuring this requires a considerable amount of adjustment by hand of the grid
points and grid size used for spanning the models state space. Further details of the solution approach are described
in online Appendix A12. The MatLab code used for solving the model is available upon request.
67
The targets are chosen to match features of the fundamental processes emphasized in the asset-pricing literature.
68
See online Appendix A4 for further details.
69
For details on how this can be achieved, see online Appendix A4.
70
Since the gain parameter g will be small,
2is also small, hence the contribution of
2to the total variance
in (26) is negligible.
where is the set of moments in the data to be matched (the ones listed in Table3),
()the corresponding moment from the model for parameter values , and a
weighting matrix.
We pursue two estimation approaches, one that implements efficient SMM and an
alternative one that emphasizes more directly the replication of the data moments.
Both approaches have advantages and disadvantages, as we discuss below, but ulti-
mately deliver similar estimates and model moments. All estimations exclude the
risk-free rate from the set of moments to be matched, as the model has a hard time in
fully replicating the equity premium. We thus only report the risk-free interest rate
implied by the estimated models.
Our first estimation approach chooses the weighting matrix in equation (44)
to be equal to the inverse of the estimated covariance matrix of the data moments
, as required for efficient SMM estimation. While an efficient weighting matrix is
desirable for estimation, it also causes some difficulties. First, the weighting matrix
turns out to be approximately singular: following Adam, Marcet, and Nicolini
(2016), we thus exclude some moments from the estimation that are nearly redun-
dant, namely we exclude the excess return regression coefficient (c) from the set of
estimated moments.71 Second, we also exclude the correlation between the PD ratio
and surveyed expected returns, since the short sample for surveys provides a less
reliable estimate of .
The second estimation approach uses a diagonal weighting matrix in equa-
tion(44), with the diagonal entries consisting of the inverse of the individually esti-
mated variances of the corresponding data moments in . Although this alternative
is less efficient from an econometric point of view, the matrix is now guaranteed
to be invertible, therefore we can use the full set of moments in the estimation.72
Furthermore, this criterion just minimizes the sum of t-statistics of all the moments,
therefore emphasizes more directly matching the moments, as typically pursued in
the asset-pricing literature.
An unconstrained minimization of the objective function (44) over =(g,,,p)
turns out to be numerically unstable and computationally too costly. For this reason,
we impose additional restrictions on the parameter space . These restrictions can
only constrain the empirical performance of the model, so that the goodness of fit
results presented below constitute a lower bound on what the model can potentially
achieve. Specifically, we impose ( D) = 0.995( D ) 2, where Dassumes the
value from Table 3. This additional restriction is inspired by the fact thataccording
to our experiencethe model can perform reasonably well for ( ,) = (0.995,2)
and helps resolving numerical instabilities in our solution routines. We furthermore
restrict the persistence parameter to p {0.95,0.999}, which is inspired by the fact
that the sample autocorrelation of log(1+Wt/Dt )is very high in US postwar data
(about 0.99).
71
See Section V and the online Appendix in Adam, Marcet, and Nicolini (2016) for details on asymptotic
d istribution of SMM, how to use a systematic criterion for excluding moments, details on how to estimate
and
how to compute ()using small samples.
72
As mentioned before, the risk-free rate is always excluded from estimation.
Table 3 reports the estimation outcome in terms of implied model moments, esti-
mated parameters, and t-ratios.73
In terms of estimated parameters, the discount factor is estimated to be close to 1
and relative risk aversion is slightly above 2. The estimated gains (g) are very close
to the values obtained in the empirical Section IID.
For both estimation approaches, the model matches the data moments rather well.
In particular, the model easily replicates the positive correlation between the PD
ratio and expected stock returns (corr[PDt ,E t Rt+1 ]), in line with the value found
in the survey data. This is achieved, even though this moment was not used in the
estimation using the efficient weighting matrix. The model also performs well in
terms of producing sufficient volatility for the PD ratio (std[PD]) and stock returns
(std[r s ]). If anything, the model tends to produce too much volatility. The model
also succeeds in replicating the mean and autocorrelation of the PD ratio (E[PD]
and corr[PDt ,PDt1]) and the evidence on excess return predictability (cand R 2),
even though the regression coefficient cwas not included in the set of moments to
be matched.
The most significant shortcoming of the model concerns the equity premium.
While it matches the average stock return (E[r s ]), it predicts a too high value for
the risk-free rate (E[r b ]). Nevertheless, the model can explain about one-half of
the equity premium observed in the data. Given the low estimated value for the
degree of relative risk aversion () , this is remarkable and we explore the mecha-
nism behind this outcome further in Section VIIIB below.
Abstracting from the mean risk-free rate, the estimation using the diagonal
weighting matrix generates t-ratios below 2 for all model moments and often even
t-ratios below 1. The estimation using the efficient weighting matrix performs sim-
ilarly well, but implies a slightly high volatility for stock returns and the PD ratio.
The empirical performance of the subjective belief model is overall very good, espe-
cially when compared to the performance of the rational expectations version of
the model, as reported in Table 4 using the same parameters as for the estimated
model (with diagonal matrix) in Table 3. Under RE, the t-ratios all increase in abso-
lute terms, with some of the increases being quite dramatic. With objective price
beliefs, the model produces insufficient asset price volatility (too low values for
std[PD] and std[r s ]) and the wrong sign for the correlation between the PD ratio
and expected stock returns (corr[PDt ,E t Rt+1 ]). It also gives rise to a small negative
equity premium.74 These features are rather robust across alternative parameteriza-
tions of the RE model and highlight the strong quantitative improvement obtained
by incorporating subjective belief dynamics.
Table 5 shows that the performance of the subjective belief model is also rather
robust across different values for the persistence parameter p of the wage-dividend
process. The table reports the estimation outcomes when repeating the e stimation
73
The t-ratio is the ratio of the gap between the model and the data moment over the standard deviation of the
moment in the data, as implied by the weighting matrix. For excluded moments we use the individually estimated
standard deviations in the numerator. For the case of an efficient weighting matrix, the t-ratio for moments included
is computed according to the proper covariance matrix that delivers a standard normal distribution (see the online
Appendix in Adam, Marcet, and Nicolini 2016).
74
For the considered parameterization, the equity premium is slightly negative because stocks are a hedge
against wage income risk.
Table 4RE Asset-Pricing Moments
US data RE model
Moment Moment t-ratio
E[PD] 139.8 108.3 1.26
std[PD] 65.2 11.3 3.64
corr[PDt, PDt1] 0.98 0.95 11.20
std[R] 8.00 1.49 16.22
c 0.0041 0.0181 11.09
R2 0.25 0.72 3.66
E[R] 1 1.89 1.41 1.04
E[R b] 1 0.13 1.52 8.15
UBS survey data

corr[PDt,Et Rt,t+4] 0.79 1.00 26.36
Notes: The table reports US asset-pricing moments (second column) using the data sources
described in online Appendix A1, the moments of the rational expectations (RE) model (col-
umn3), and the t-ratios of the RE model (column 4). The RE model uses the parameterization of
the estimated subjective belief model from Table 3 (diagonal matrix). See Table 3 for a descrip-
tion of the moment labels in the first column.
Table 5Asset-Pricing Moments (Restricted Estimation)
US data Subjective Subjective

1946:I2012:I belief model belief model
(quart. real values) (diagonal matrix) (diagonal matrix)
Restriction imposed p=0 p=0.999
E[PD] 139.8 112.9 1.07 116.8 0.92
std[PD] 65.2 82.1 1.14 98.5 2.25
corr[PDt, PDt1] 0.98 0.98 0.22 0.99 1.78
std[R] 8.00 7.91 0.22 7.73 0.68
c 0.0041 0.0045 0.36 0.0037 0.27
R2 0.25 0.15 0.74 0.14 0.79
E[R] 1 1.89 1.83 0.14 1.81 0.17
E[R b] 1 0.13 0.97 4.92 1.00 5.12
UBS survey data

corr[PDt,Et Rt,t+4] 0.79 0.85 0.82 0.75 0.65
Estimates
g 0.0222 0.0272
0.99510 0.99519
2.02 2.04
AppendixA1, the moments and t-ratios of the estimated subjective belief model when restricting the persistence
parameter p from equation (16) to zero (columns 3 and 4) and when restricting p to 0.999 (columns 5 and 6). See
Table 3 for a description of the labels used in the first column.
from Table 3 but imposing p=0 (no persistence) and p=0.999 (near unit
root behavior). The ability of the subjective belief model to match the empirical
asset-pricing moments is hardly affected by the persistence parameter.
Figure 7 illustrates how the subjective belief model improves empirical perfor-
mance. The figure depicts the equilibrium PD ratio (y-axis) as a function of agents
capital gain beliefs (x-axis). It graphs this relationship once for the model with uncer-
tainty (dark line) and once for the vanishing noise limit analyzed in the p revious
450
400
350
Quarterly PD ratio
300
250
200
150
100 PD ratio without uncertainty

PD ratio with uncertainty
50
0.03 0.02 0.01 0 0.01 0.02 0.03 0.04 0.05
Subjective price growth expectations (mt 1)
Figure 7. The Effects of Uncertainty on the Equilibrium PD Ratio

(Estimated Model from Table 3, Diagonal Matrix)
section (lighter line).75 While the presence of price, dividend and wage risk lowers
the equilibrium PD ratio compared to a setting without risk, the functional form of
the relationship remains qualitatively unchanged. The intuition from the vanishing
noise limit thus carries over to the model with noise: the model continues to give rise
to occasional boom and bust dynamics in asset prices.
A. Further Evidence on Model Performance
Table 6 presents further asset-pricing moments that have not been used in the
estimation of the model. It reports data moments, the corresponding model moments
for the estimated subjective belief model from Table 3, for the RE version of model
and for an augmented subjective belief model featuring stock supply shocks, as
introduced below.
Table 6 shows that the subjective belief model successfully replicates the low
volatility of the risk-free interest rate present in the data. The subjective belief model
even sightly underpredicts volatility. This shows that the ability to generate highly
volatile stock returns does not rely on counterfactually making the risk-free interest
rate very volatile. In terms of matching the low volatility of the risk-free rate, the
performance of the subjective belief model is approximately as good as that of the
RE model.
Table 6 also presents evidence on the quarterly autocorrelations of stock returns
and excess stock returns. It shows that the subjective belief model overpredicts both
autocorrelations relative to the ones present in the data. As Table 6 shows, this feature
75
For the stochastic solution, the equilibrium PD in Figure 7 is determined from the market-clearing condition
(38) assuming W t /Dt = , to be comparable with the value this variable assumes in the vanishing risk limit. The
figure assumes the parameters implied by the estimated model with diagonal matrix from Table 3.
Table 6Further Data and Model Moments
Estimated model RE version estimated

(Table 3, (Table 3, diagonal Estimated model
diagonal matrix) matrix) with supply shocks
Data Standard
moment deviation Moment t-ratio Moment t-ratio Moment t-ratio
Standard deviation 1.11 0.24 0.83 1.16 0.62 2.04 0.62 2.05
risk-free rate
Autocorrelations
Excess stock return 0.12 0.046 0.53 9.08 0.33 4.78 0.12 0.18
Stock return 0.13 0.045 0.54 9.21 0.04 1.87 0.14 0.37
arises not solely due to the presence of subjective beliefs: the rational e xpectations
versions of the model also overpredicts the autocorrelation of excess stock returns.
The failure of the subjective belief model to deliver a near zero high-frequency
autocorrelation for returns is related to the fact that the low frequency returns gen-
erated by learning also show up at high frequency because the model features few
transitory shocks to stock prices. As we show below, the subjective belief model
becomes consistent with the observed value of the quarterly autocorrelations of
stock returns and excess stock returns, when adding small i.i.d. disturbances to the
model.
To illustrate this point in a simple way, we consider in online Appendix A13 a
model where equilibrium stock supply in period t satisfies
St =e t ,
s

where st iiN( 2/2,

s
2s )denotes an exogenous shock to the supply of stocks.
This shock can represent extraneous demand for stocks, e.g., the one arising from
noise or liquidity traders.76 This generalized setup nests the model without noise
traders for the special case S = 0.
As it turns out, small positive values for the standard deviation Sbring the
model fully in line with the observed autocorrelations. The last two columns in
Table 6 illustrate this fact and report the model implied autocorrelations, when set-
s = 1.2510 3, which implies that the standard deviation of aggregate stock
ting
supply is 0.125 percent of the average outstanding amount of stocks (which is equal
to 1). Table 6 shows that the autocorrelations are now fully consistent with those
observed in the data, while the standard deviation of the risk-free rate is hardly
affected. Moreover, as shown in online Appendix A13, the presence of stock sup-
ply shocks has only small effects on the other asset-pricing moments reported in
Table3.
76
Alternatively, the shocks St may capture changes to asset float, as discussed in Ofek and Richardson (2003)
and Hong, Scheinkman, and Xiong (2006). In any case, they capture (exogenous) stock demand or supply that is
not coming from the consumers described in equation (14).
B. Subjective Consumption Plans and the Sharpe Ratio
A somewhat surprising feature of the estimated models in Table 3 is that they

give rise to a fairly high Sharpe ratio, despite the fact that the estimated risk aver-
sion is relatively low. Specifically, the estimated models imply that the quarterly
unconditional Sharpe ratio, ( E[r s ]E[r b ])/std[r s ],is close to 0 .1. Since the model
implied equity premium is too low, this value falls significantly short of the value
of Sharpe ratio in the data (0.22). Yet, it significantly exceeds the value it would
assume if agents held rational price expectations. Under RE, the Sharpe ratio would
be tiny and approximately be given by77
E[R]E[R b ]
(45) __________
E[stdt (Ct+1/Ct)] = 0.00576.

std[R]
In the presence of subjective price beliefs, however, the Sharpe ratio is approxi-
mately equal to
[ t] [ t t+1 ]
E R E R b
[ t+1] [ t+1] E R E E [R ]
(46)____________
________________

std[R] std[R]

Subj. return pessimism
[ t t+1 t ]
E std [C /C]
+ ______________

E[stdt[Ct+1/ Ct]],

E[stdt[Ct+1
/Ct]]

Ratio of

subj. to obj.
consumption volatility
which illustrates the existence of two additional factors affecting the Sharpe ratio:
subjective return pessimism, as present in models where agents entertain robustness
concerns, e.g., Cogley and Sargent (2008), and the relative volatility of subjective
consumption plans compared to the ones emerging under rational expectations.
Subjective return pessimism contributes to the Sharpe ratio because pessimism
depresses stock prices and thereby generates a higher equity premium expost.
Subjective consumption volatility affects the equity premium for standard reasons.
For the estimated models from Table 3, we find that about one-third of the
models Sharpe ratio is due to subjective return pessimism and about two-thirds due
to the second term on the RHS of equation (46).78 In both estimated models, the
second component is so large because the average subjective standard deviation of
consumption growth is about 12.5 times larger than the average objective standard
deviation.79
77
See online Appendix A14 for a derivation of equations (45) and (46).
78
The approximate Sharpe ratio decomposition (46) works quantitatively well for the estimated models from
Table 3: the actual Sharpe ratios are 10.7 percent and 10.2 percent for the model estimated using a diagonal and an
efficient matrix, respectively, while the terms on the RHS in equation (46) sum to 10.5 percent and 10.9 percent.
79
The precise ratios are 1 2.87(12.64) for the model from Table 3 estimated using a diagonal (efficient) matrix.
For both models, we have E[stdt [Ct+1
/Ct ]] = 0.00273.
Analyzing which future shocks contribute in agents mind to this high level of sub-
jectively expected consumption volatility, we find that transitory price growth shocks
are the main culprit, i.e., the perceived shocks t+1in the subjective price evolution
equation (26). These shocks move prices permanently up (or down), without an asso-
ciated change in fundamentals, and prompt agents to make contingent plans to sell (or
buy) stocks in the future in response to these shocks. These expectations of future trade
increase subjective expected consumption volatility, while the representative agent
assumption insures that agents will never actually buy or sell stocks in equilibrium. As
a result, subjective and objective consumption volatility diverge considerably.
The reason for this discrepancy is largely due to the fact that the homogeneous
agent model imposes zero trade in equilibrium. Equilibrium consumption is then not
affected by stock prices. As we now illustrate, one can resolve this discrepancy by
introducing heterogeneity among agents.
We show this using two simple heterogeneous agent extensions of our baseline
model. The first extension studies an ad hoc modification in which agents beliefs
at time tare given by
(47) lnmti = lnmt + ln it ,
where ln it iiN(0, 2) is an idiosyncratic and transitory shock to agent is capital

gain optimism and where m tevolves as in the baseline model. The second extension
considers a fully microfounded extension, wherein line with the belief updating
equation (39)agents expectations evolve according to
+ g(lnPt1 lnPt2 lnmti1

(48) lnmti = lnmti1 ) + gln 1,i
t ,
where ( gln 1,i t )

iiN(0, 2,1)is idiosyncratic with rather persistent effects on
beliefs.80 In both extensions, belief heterogeneity induces trade in equilibrium,
which increases the objective volatility of individual consumption growth relative
to the representative agent setting. We then calibrate the variances 2 and 2,1such
that the objective and subjective standard deviations are approximately aligned
(E[ std t [Ct i+1
/Ct i ]] E[stdt[Ct i+1
/Ct i ]] ).
Table 7 reports the asset-pricing moments and t-ratios for both heterogeneous
agent extensions.81 As becomes clear from Table 7, the extended models can rep-
licate the asset-pricing moments equally well as the baseline model, but also align
the subjective and objective standard deviations of consumption growth. The ability
of the model to generate a low risk-free rate is actually improved relative to the
representative agent model, with the average risk-free rate even becoming slightly
negative for the model with persistent shocks.82 Overall, Table 7 illustrates the
80
As is clear from online Appendix A7, the shock ln 1,i
t represents information that the agent receives about
the perceived transitory price shock. Since these shocks exist only in subjective terms, the baseline model sets
ln t = lnt = 0for all t.
i
81
The table uses the model parameterization of the estimated model from Table 3 (diagonal matrix). For the
case with persistent belief shocks, we slightly adjust the updating gain gto improve the match with the asset-pricing
moments. The simulations use 51 different agents.
82
The strong effect on the risk-free rate is due to the tight borrowing limits (Bt i 0) that we impose: see
the discussion in Section III. The effect would likely be less strong for less strict borrowing limits, but this would
require fundamental changes to the solution approach, including an increase in the number of state variables.
Table 7Heterogeneous Agent Model with Equilibrium Trade
US data Subjective belief model Subjective belief

1946:I2012:I (transitory belief model (persistent
(quart. real values) shocks) belief shocks)
E[PD] 139.8 113.5 1.05 120.6 0.77

std[PD] 65.2 85.9 1.40 90.8 1.73
corr[PDt , PDt1] 0.98 0.98 0.63 0.98 0.51
std[R] 8.00 7.37 1.58 7.93 0.16
c 0.0041 0.0047 0.53 0.0064 1.81
R2 0.25 0.18 0.47 0.25 0.05
E[R] 1 1.89 1.80 0.19 1.87 0.06
E[R b] 1 0.13 0.34 1.26 0.65 4.58
UBS survey data

corr[PDt,Et Rt,t+4] 0.79 0.79 0.08 0.85 0.78
Consumptive volatility
E[std t Ct i+1
/ Ct i ] 3.38 4.55
/ Ct i ]
E[stdt Ct i+1 3.23 4.99
Estimates
g 0.0282 0.0282
0.99514 0.99514
2.03 2.03
0.0030
,1 0.0018
AppendixA1, the moments and t-ratios of the subjective belief model with idiosyncratic transitory belief shocks
(columns 3 and 4, see equation (47)), and the moments and t-ratios of the subjective belief model with idiosyncratic
persistent belief shocks (columns 5 and 6, see equation (48)). E[std t Ct+1/Ct] denotes the unconditional expectation
of the subjective standard deviation of consumption growth and E[stdt Ct+1 /Ct] the unconditional expectation of the
objective standard deviation of consumption growth. The standard deviation of the idiosyncratic belief shocks
and has been chosen so as to approximately equate subjective and objective standard deviation of consumption
growth. See Table 3 for a description of the remaining labels used in the first column.
robustness of our quantitative results toward allowing for investor heterogeneity.

Adam et al. (2015) explore in greater detail a heterogeneous agent version of the
present model and show that the model successfully matches important stylized
facts regarding trading volume.
C. Expectational Errors
Since our model matches stock price behavior and the comovement pattern
between survey expectations and stock prices, agents expectations in the model
fail to satisfy the rational expectations hypothesis, in line with what has been doc-
umented for the survey evidence. To document the quantitative importance of this
fact, this section applies the RE tests from Section IIB to the subjectively expected
one-quarter-ahead capital gain expectations from our estimated models. In particu-
lar, we test the acceptance rates of the rational expectations hypothesis (H0:c=c)
at various significance intervals and various sample lengths, using the test statistic
Table 8Rejection Frequencies for the RE Hypothesis (H0: c=c)

for the Estimated Asset-Pricing Models
Significance Sample length in quarters

level 25 50 75 100 150 200
Model from Table 7 1% 0.828 0.944 0.971 0.981 0.984 0.987
(diagonal matrix) 5% 0.900 0.970 0.981 0.987 0.988 0.990
Model from Table 7 1% 0.779 0.938 0.968 0.975 0.975 0.989
(efficient matrix) 5% 0.892 0.974 0.986 0.981 0.980 0.991
Note: The reported rejection frequencies are based on the rational expectations test developed
in Section IIB of the paper and include the small sample bias correction stated in Proposition 1.
described in Proposition 1.83 Table 8 reports the outcomes.84 It shows that at con-
ventional significance levels the rational expectations hypothesis can be rejected
with high likelihood, even for relatively short sample lengths. This is in line with the
empirical evidence documented in Section II.
IX. Historical PD Ratio and Survey Evidence
This section shows that the estimated subjective belief model successfully rep-
licates the low-frequency movements of the postwar US PD ratio, as well as the
available time series of survey expectations. We illustrate this point using the model
from Table 5, which has been estimated under the restriction that the wage-dividend
ratio displays close to unit root behavior (p=0.999). This is motivated by the
fact that estimated models that imply more mean reversion in the wage-dividend
ratio, say the ones from Table 3, cannot capture the time series behavior of the PD
ratio after the year 1990. After 1990, the wage-dividend ratio started to persistently
diverge from a value close to its sample mean of 22 to a value close to 12 at the end
of the sample period in 2012.
To determine the model fit, we feed the historically observed price growth observa-
tions into the models belief updating equation (39), so as to obtain a model-implied
belief process.85 We then combine this belief series with the historically observed
wage-dividend ratio to obtain from the models equilibrium pricing function the
model implied prediction for the postwar PD series. Formally, this is done by plug-
ging implied expectations and wage-dividend ratio in equation (38) and backing out
the corresponding PDratio. In doing so, we choose the unobserved price growth
belief at the start of the sample period (m1946:I) so as to minimize the sum of absolute
deviations of the model-implied PD ratio from the PD ratio in the data.86
83
We include the small sample bias correction reported in Proposition 1.
84
Rejection frequencies have been computed from 1,000 random samples of the specified sample length that
were randomly drawn randomly from a time series of 100,000 simulated model periods, with the first 10,000
periods being discarded as burn-in.
Since we cannot observe the shocks ln 1t in (39), we set them equal to 0. Gaps between the model predicted
85
and actual PD ratio may thus partly be due to these shocks.

This delivers initial price growth expectations in 1946:I of 1.7 percent.
86
0.03
Subjective price growth expectations (m(t) 1)

0.02
0.01
0.01
0.02
0.03
1940 1950 1960 1970 1980 1990 2000 2010 2020
Figure 8. Price Growth Expectations Implied by Bayesian Updating

and Historical Stock Price Growth
(Estimated Model from Table 5, p=0.999), 1946:I2012:I
The resulting belief sequence is depicted in Figure 8. Figure 9 graphs the model
implied PD ratio together with the PD ratio from the data. It reveals that the model
captures a lot of the low-frequency variation in the historically observed PD ratio. It
captures particularly well the variations before the year 2000, including the strong
run-up in the PD ratio from the mid 1990s to the year 2000. The model also predicts
a strong decline of the PD ratio after the year 2000, but overpredicts the decline
relative to the data.
The gap after the year 2000 emerging in Figure 9 is hardly surprising, given the
empirical evidence presented in Figure 4, which shows that the relationship between
the PD ratio and the expectations implied by equation (13) has shifted upward in
the data following the year 2000. While we can only speculate about potential rea-
sons causing this shift, the exceptionally low real interest rates implemented by the
Federal Reserve following the reversal of the tech stock boom and following the
collapse of the subsequent housing boom may partly contribute to the observed
discrepancy. This suggests that incorporating the asset-pricing effects of monetary
policy decisions might improve the model fit. This is, however, beyond the scope of
the present paper.
Figure 10 depicts the model-implied price growth expectations and the price
growth expectations from the UBS survey.87 While the model fits the survey data
rather well, it predicts after the year 2003 considerably lower capital gains expecta-
tions, which partly explains why the model underpredicts the PD ratio in Figure9
toward the end of the sample period. While the expectations gap in Figure10 nar-
rows considerably after the year 2004, this fails to be the case for the PD ratio in
87
See footnote 35 for how to compute price growth expectations from the UBS survey.
400
Model implied PD ratio
PD ratio S&P 500
350
300
Quarterly PD ratio
250
200
150
100
50
1940 1950 1960 1970 1980 1990 2000 2010 2020
Figure 9. Historical PD Ratio: Data versus Model, 1946:I2012:I

(Estimated Model from Table 5, p=0.999)
Subjective price growth expectations (m(t)1)
0.035
Price growth expectations implied by Bayesian updating
UBS expected price growth (own portfolio)
0.03
0.025
0.02
0.015
0.01
0.005
0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Figure 10. Price Growth Expectations: UBS Survey versus Bayesian Updating Model
(Estimated Model from Table 5, p=0.999), 1998:II2007:IV
Figure 9. Underprediction of expected price growth thus explains only partly the
deterioration of the fit of the PD ratio toward the end of the sample period.
X. Robustness Analysis
This section explores the robustness of our main findings to changing key model
parameters and to using more general price belief systems. The next section stud-
ies the pricing effects of alternative model parameters. Section XB discusses the
implications of more general price belief systems that incorporate mean-reversion
of the PD ratio. Finally, Section XC considers the pricing and welfare implications
of introducing agents who use current price information for updating beliefs.
Panel A
Quarterly PD ratio
1,000
= 2 (baseline)
= 2.5
= 1.5
500
0
0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 0.1 0.12
Panel B
Quarterly PD ratio
600
= 22 (baseline)
400 = 26
= 18
200
0
0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 0.1 0.12
Panel C
Quarterly PD ratio
1,000
= 0.995 (baseline)
= 0.998
500 = 0.992
0
0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 0.1 0.12
Figure 11. Effects of (,,) on the Equilibrium PD Function
A. Model Parameterization
This section studies how the models ability to generate boom and bust dynamics
in stock prices depends on key model parameters. The equilibrium pricing function
for our baseline parameterization, depicted in Figure 5, allows for self-reinforcing
stock price boom and bust dynamics because the price dividend ratio increases ini-
tially strongly with capital gain optimism. This section explores the robustness of
this feature by studying how the equilibrium pricing is affected by the coefficient of
relative risk aversion (), the discount factor (), and the average wage to dividend
income ratio ().
Figure 11 depicts the equilibrium pricing function for alternative parameter
choices. Each panel plots the pricing function from our baseline parameterization,
as well as those generated by increasing or decreasing the values for , , and
.
Panel A, for example, shows that lowering (increasing) the coefficient of relative
risk aversion increases (reduces) the hump in the PD function and moves it to the
left (right), thereby causing asset price booms to become more (less) likely and
larger in size. Similar effects are associated with increasing (decreasing) the wage to
dividend income ratio () and with increasing (decreasing) the discount factor ().
Overall, Figure 11 shows that the model can produce hump-shaped equilibrium PD
functions over a fairly wide set of parameter specifications.
B. Generalized Belief System
This section considers a generalized price belief system, which implies that
investors expect the PD ratio to eventually mean-revert over time. This is motivated
by the fact that the price belief system (26)(27) employed in the main part of the
paper, along with knowledge of the dividend process, does not imply that agents
expect the log of the P Dratio to mean revert over time.
We find that the pricing implications of the model do not depend on the expected
long-run behavior for the PD ratio and that very similar equilibrium pricing func-
tions can be generated by belief systems that imply persistent changes in the PD
ratio but where the PD ratio is ultimately mean reverting.
It is important to note that the survey data provide little support for mean-revert-
ing price growth expectations over the available forecast horizons and thus for mean
reversion in the expected PD ratio. In particular, the last four rows in Table 1A show
obtained from regressing the expected 10 year ahead
that the regression coefficient c
capital gain forecasts from the Shiller survey on the PD ratio is more than 10times
larger than the corresponding regression coefficient obtained from regressing the
one year ahead forecast on the PD ratio. The expected annualized ten year price
growth thus reacts stronger to movements in the PD ratio than the expected one year
price growth. This shows that one should incorporate only a mild degree of mean
reversion into subjective price growth beliefs, as the belief system would otherwise
become inconsistent with the survey evidence.
To analyze the effects of mean reversion in price beliefs, we now assume that
investors perceive prices to evolve according to
lnPt+1 = lnt+1 + lnPt +(1PD)(lnPDlnPt /Dt) + lnt+1,

(49)
(50) lnt +1 = (1 )ln D + lnt + lnt+1 ,
where lnPDdenotes the perceived long-run mean of the log PD ratio

and P D [ 0,1], [ 0,1]are given parameters. For P D = = 1 these
equations deliver the benchmark price belief system (26)(27) studied in the main
part of the paper.88 For < 1 , equation (50) implies that agents expect mean
reversion in the persistent price growth component lnttoward the mean growth
rate of dividends (ln D); if in addition PD < 1 , equation (49) implies that agents
expect lnPt /Dtto eventually return to its long-run mean lnPD.89
Suppose that < 1and PD < 1, that agents are optimistic about future cap-
ital gains, i.e., they believe t to be above D, and that agents observe a PD ratio
above its long-run mean (Pt /Dt > PD). Provided and PD are sufficiently close
to 1, equations (49)(50) imply that agents expect a fairly persistent boom in the PD
ratio, as is the case with the benchmark belief system (26)(27). Yet, unlike with the
88
As before, we assume that agents have rational expectations about the dividend and wage income processes.
89
This can be seen by subtracting (15) from (49).
benchmark, they also expect a price bust further down the road, because the PD ratio
is expected to eventually return to its long-run value.
Equations (49) and (50) jointly imply that optimal belief updating about
the unobserved persistent stock price growth component lntis described by
a generalized version of equation (39), which states that the posterior mean
lnmt E (lnt |P t )in steady state evolves according to90
(51)
lnmt = (1 )ln D + lnmt1

lnPt1 lnPt2 lnmt1

((1PD )(lnPDlnPt2 /Dt2 ))
g
+ + gln 1t .

Using the generalized belief and updating equations, Figure 12 depicts the impact of
different perceived values for P Dand on the equilibrium pricing function when
setting the perceived long-run mean lnPD in (49) equal to the perfect foresight
value of the PD ratio.91 Panel A plots the effects of decreasing P Dbelow one, while
keeping = 0.9999. For the considered values of PD, agents expect the log PD
ratio to mean revert by 1 percent, 2 percent, or 3 percent per year toward its long-run
mean (lnPD). The figure also plots the outcome when there is virtually no mean
reversion (P D = 0.9999). Panel A shows that by introducing mean reversion, one
pushes the peak of the equilibrium PD ratio to the right and also lowers its height.
The shift to the right occurs because agents only expect a persistent boom in the
PD ratio if the increase in the PD ratio implied by the persistent growth component
(lnt )outweighs the mean reversion generated by the negative feedback from the
deviation of the PD from its long-run value in equation (49). The downward shift
in the PD ratio occurs because mean reversion causes investors to expect lower and
eventually negative returns sooner.
Panel B in Figure 12 depicts the effects of decreasing below 1, while keep-
ing P D = 0.9999. As before, we consider values for that imply virtually no
mean reversion and mean reversion by 1 percent, 2 percent, and 3 percent per year
toward the long-run value (ln D ). The panel shows that the pricing implications are
very similar to those of decreasing P D.
Finally, panel C depicts the effects of jointly decreasing and P D. The pricing
implication of such simultaneous changes turn out to be considerably stronger, com-
pared to the case where only one of the persistence parameters decreases. Figure 13
illustrates why this sharp difference occurs. It graphs the expected path of the PD
ratio for different parameter combinations ( PD ,) .92 If either PD
or are close
to 1, agents expect a rather prolonged stock price boom that is expected to revert
only in the distant future. Expected stock returns are thus high for many periods
Since equation (49) only introduces an additional observable variable into (26) and equation (50) only adds
90
a known constant and known mean reversion coefficient relative to (27), the arguments delivering equation (39) as
optimal Bayesian updating directly generalize to equation (51).
We assume that 2and 2assume the same values as in the baseline specification, so that the gain parameter
91
also remains unchanged at g=0.02515. This parameterization makes sure thatfor the values of and PD
considered belowthe perceived standard deviation for stock price growth implied by equations (50)(49) approx-
imately matches the standard deviation of stock price growth in the data.
The figure assumes that the equilibrium PD ratio initially equals PD0 = 150, i.e., it is above its long-run
92
value, and that agents are mildly optimistic about future capital gains with lnm0 = 1%>ln D.
Panel A
600
Quarterly PD ratio
PD = 0.9999
400 PD = 0.991/4
PD = 0.981/4
200 PD = 0.971/4
0
0.1 0.05 0 0.05 0.1 0.15
Panel B
600
= 0.9999
Quarterly PD ratio
400 = 0.991/4
= 0.981/4
200 = 0.971/4
0.1 0.05 0 0.05 0.1 0.15

Panel C
600
PD = = 0.9999
Quarterly PD ratio
400 PD = = 0.991/4
PD = = 0.981/4
200 PD = = 0.971/4
0.1 0.05 0 0.05 0.1 0.15

Figure 12. Equilibrium Pricing Function (Generalized Belief System)
300
250
200
Quarterly PD ratio
150
100
PD = 0.9999, = 0.971/4
PD = 0.971/4, = 0.9999
50
PD = 0.971/4, = 0.971/4
0
0 100 200 300 400 500 600 700 800 900 1,000
Figure 13. Persistence Parameters and Expected PD Path (Generalized Belief System)
before they turn negative. This differs notably from the case where both persistence
_
parameters fall significantly below one (PD = = 0.97 4). The expected stock
1
price boom is then much smaller and considerably more short-lived, so that returns
are lower and expected to become negative earlier. From the discussion following
equation (42), it should be clear that the implied path for expected returns then can-
not sustain a high PD ratio as an equilibrium outcome.
Summing up, the model continues to give rise to hump-shaped equilibrium pric-
ing functions, even if agents ultimately expect mean reversion in the PD ratio, pro-
vided the generalized belief system implies that agents expect elevated capital gains
to be sufficiently persistent.
C. Current Price Information for Belief Updating
The baseline model specification postulates a belief structure that makes it sub-
jectively optimal for agents to update beliefs mtbased on price information up to
period t1only, see equation (39). The present section demonstrates that lagged
belief updating is also objectively optimal for agents: belief updating with current
price observations robustly generates lower experienced utility for agents.93 In this
sense, the baseline specification with lagged belief updating can be interpreted as
the limiting outcome of a setting in which agents choosebased on experienced
utilitywhether to use current prices for updating beliefs.
To show that lagged updating generates objectively higher utility, we consider an
extended model setup where a share [0,1]of investors updates beliefs using
current price growth information, with the remaining share 1using lagged price
growth information.94 While the beliefs of lagged updaters are predetermined, the
beliefs of current updaters vary simultaneously with the stock price Pt. The latter
creates a potential for multiple market-clearing equilibrium price and belief pairs.
To assess the implications of simultaneous belief updating, we consider the esti-
mated benchmark parameters from Table 3 (diagonal matrix) and different values
for [0,1]. To select between equilibrium prices in periods where multiple
equilibrium price exist, we consider two alternative selection rules.95 The first rule
selects the equilibrium price that is closest to the previous periods market-clearing
price. The second selection rule selects the price that is furthest away from the pre-
vious market-clearing price.
Table 9 reports the expected discounted utility of agents that use current and
lagged belief updating for different values of and for the two considered selection
rules.96 It shows that independently of the selection rule and independently of the
share of current updaters , utility of lagged updaters always exceeds that of current
93
This finding is in line with results reported in Adam et al. (2015), who show that agents whose beliefs are
more reactive to price growth observations tend to do worse than agents whose beliefs display less sensitivity.
94
Updaters using current price information update beliefs according to equation (39), but replace lnPt1
/lnPt2

by lnPt /lnPt1
on the right-hand side.
95
In the vast majority of cases, we find three market-clearing equilibrium prices, conditional on there being
multiplicity. In less than 0.1 percent of the periods with multiplicities we find five market-clearing prices.
96
The table reports the unconditional expectation of discounted consumption utility using the objective dis-
tribution for consumption, as realized in equilibrium. Online Appendix A15 reports the associated asset-pricing
moments.
Table 9Expected Utility of Current and Lagged Belief Updaters
Closest price Furthest price

Share of current Lagged Current Lagged Current
updaters () updaters updaters updaters updaters
0.0 4.02 4.07 4.02 4.07

0.1 4.02 4.06 4.02 4.06
0.2 4.01 4.06 4.01 4.06
0.3 4.01 4.05 4.01 4.06
0.4 4.01 4.05 4.00 4.06
0.5 4.00 4.05 3.99 4.06
0.6 4.00 4.04 3.98 4.06
0.7 3.99 4.04 3.97 4.06
0.8 3.99 4.03 3.89 4.11
0.9 3.97 4.03 3.39 4.24
1.0 3.95 4.02 2.46 4.02
Notes: The table reports the objectively expected utility for agents that use current and one period
lagged prices to update beliefs for different equilibrium selection rules, as described in Section
XC, and different shares of current updaters. Since u(C)<0, expected utility is always nega-
tive. The model is parameterized using the estimated parameters from Table 3 (diagonal matrix).
updaters. Current updaters thus have an incentive to switch to lagged updating, i.e.,
to the setting considered in our baseline specification.
XI.Conclusions
We present a model with rationally investing agents that gives rise to market
failures in the sense that the equilibrium stock price deviates from its fundamental
value. These deviations take the form of asset price boom and bust cycles that are
fueled by the belief-updating dynamics of investors who behave optimally given their
imperfect knowledge about the behavior of stock prices. Optimal belief updating also
causes investors subjective capital gain expectations to comove positively with the
price-dividend ratio, consistent with the evidence available from investor surveys.
As we argue, these features cannot be replicated within asset-pricing models that
impose rational price expectations. Moreover, the developed statistical tests show
that the behavior of survey return expectations is incompatible with the rational
expectations hypothesis.
Taken together, this suggests that asset price dynamics are to a large extent influ-
enced by investors subjective optimism and pessimism, i.e., the asset price fluctu-
ations observed in the data are to a large extent inefficient. The inefficiency arises
because equilibrium stock prices are determined by the sum of dividends that is
discounted using the stochastic discount factor implied by investors subjective con-
sumption plans, which are influenced by investors subjective price beliefs. This
differs from the standard setup under RE, where stock prices equal the sum of divi-
dends discounted by the objective stochastic discount factor.
Due to the simplicity of the setup, these inefficient price fluctuations do not yield
adverse welfare implications in our baseline model.97 For models incorporating
investor heterogeneity, e.g., the extension presented in Section VIIIB, or models
97
This is true if one evaluates welfare using expost realized consumption.
featuring endogenous output or stock supply processes, stock price fluctuations may
have significant effects on welfare. Exploring these within a setting that generates
quantitatively credible amounts of asset price fluctuations appears to be an interest-
ing avenue for further research. Such research will in turn lead to further important
questions, e.g., whether policy can and should intervene with the objective to stabi-
lize asset prices.
In deriving our results, we assumed that all agents in the economy become more
(or less) optimistic when observing capital gains above (or below) their expecta-
tions. While the quantitative model predictions survive when investors are heteroge-
neous in the degree to which they respond to observed capital gains (see Adam etal.
2015), it appears of interest to assess the potential price impact generated by spec-
ulators with rational price expectations. While rational speculators can contribute
to price destabilization, as in De Long et al. (1990), they may also help with price
stabilization, as in Barberis et al. (2015). Exploring this issue further, especially in
connection with the limits to arbitrage emphasized in Shleifer and Vishny (1997),
appears to be a fruitful avenue for further research.
References
Adam, Klaus, Johannes Beutel, Albert Marcet, and Sebastian Merkel. 2015. Can a Financial Transac-
tion Tax Prevent Stock Price Booms? Journal of Monetary Economics 76 (S): S90109.
Adam, Klaus, and Albert Marcet. 2010. Booms and Busts in Asset Prices. Bank of Japan Institute for
Monetary and Economic Studies Discussion Paper 2010-E-2.
Adam, Klaus, and Albert Marcet. 2011. Internal Rationality, Imperfect Market Knowledge and Asset
Prices. Journal of Economic Theory 146 (3): 122452.
Adam, Klaus, Albert Marcet, and Johannes Beutel. 2017. Stock Price Booms and Expected Capital
Gains: Dataset. American Economic Review. https://doi.org/10.1257/aer.20140205.
Adam, Klaus, Albert Marcet, and Juan Pablo Nicolini. 2016. Stock Market Volatility and Learning.
Journal of Finance 71 (1): 3382.
Angeletos, Geroge-Marios, and Jennifer LaO. 2013. Sentiments. Econometrica 81 (2): 73979.
Aruoba, S. Boragan, Jess Fernndez-Villaverde, and Juan F. Rubio-Ramrez. 2006. Comparing
Solution Methods for Dynamic Equilibrium Economies. Journal of Economic Dynamics and Con-
trol 30 (12): 24772508.
Asparouhova, Elena, Peter Bossaerts, Nilanjan Roy, and William Zame. 2016. Lucas in the Labora-
tory. Journal of Finance 71 (6): 272780.
Bacchetta, Philippe, Elmar Mertens, and Eric Van Wincoop. 2009. Predictability in Financial Mar-
kets: What Do Survey Expectations Tell Us? Journal of International Money and Finance 28 (3):
40626.
Bansal, Ravi, and Amir Yaron. 2004. Risks for the Long Run: A Potential Resolution of Asset Pricing
Puzzles. Journal of Finance 59 (4): 14811509.
Barberis, Nicholas, Robin Greenwood, Lawrence Jin, and Andrei Shleifer. 2015. X-CAPM: An
Extrapolative Capital Asset Pricing Model. Journal of Financial Economics 115 (1): 124.
Barberis, Nicholas, Andrei Shleifer, and Robert Vishny. 1998. A Model of Investor Sentiment. Jour-
nal of Financial Economics 49 (3): 30743.
Boswijk, H. Peter, Cars H. Hommes, and Sebastiano Manzan. 2007. Behavioral Heterogeneity in
Stock Prices. Journal of Economic Dynamics and Control 31 (6): 193870.
Branch, William A., and George W. Evans. 2011. Learning about Risk and Return: A Simple Model of
Bubbles and Crashes. American Economic Journal: Macroeconomics 3 (3): 15991.
Brock, William A., and Cars H. Hommes. 1998. Heterogeneous Beliefs and Routes to Chaos in a Sim-
ple Asset Pricing Model. Journal of Economic Dynamics and Control 22 (89): 123574.
Bullard, James, and John Duffy. 2001. Learning and Excess Volatility. Macroeconomic Dynamics
5 (2): 272302.
Campbell, John Y. 2003. Consumption-Based Asset Pricing. In Handbook of the Economics of
Finance, Vol. 1B, edited by George M. Constantinides, Milton Harris, and Ren Stulz, 80387.
Amsterdam: Elsevier.
Campbell, John Y., and John H. Cochrane. 1999. By Force of Habit: A Consumption-Based Explana-
tion of Aggregate Stock Market Behavior. Journal of Political Economy 107 (2): 20551.
Campbell, John Y., and Motohiro Yogo. 2006. Efficient Tests of Stock Return Predictability. Journal
of Financial Economics 81 (1): 2760.
Carroll, Christopher D. 2006. The Method of Endogenous Gridpoints for Solving Dynamic Stochas-
tic Optimization Problems. Economics Letters 91 (3): 31220.
Choi, James J., and Thomas M. Mertens. 2013. Extrapolative Expectations and the Equity Premium.
https://c8475b0a-a-62cb3a1a-s-sites.googlegroups.com/site/thomasmichaelmertens/papers/
Choi%20Mertens%20-%20Extrapolative%20Expectations.pdf (accessed June 12, 2017).
Cogley, Timothy, and Thomas J. Sargent. 2008. The Market Price of Risk and the Equity Premium: A
Legacy of the Great Depression? Journal of Monetary Economics 55 (3): 45476.
Collin-Dufresne, Pierre, Michael Johannes, and Lars A. Lochstoer. 2016. Parameter Learning in Gen-
eral Equilibrium: The Asset Pricing Implications. American Economic Review 106 (3): 66498.
De Long, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann. 1990. Pos-
itive Feedback Investment Strategies and Destabilizing Rational Speculation. Journal of Finance
45 (2): 37995.
Epstein, Larry G., Emmanuel Farhi, and Tomasz Strzalecki. 2014. How Much Would You Pay to
Resolve Long-Run Risk? American Economic Review 104 (9): 268097.
Eusepi, Stefano, and Bruce Preston. 2011. Expectations, Learning, and Business Cycle Fluctuations.
American Economic Review 101 (6): 284472.
Evans, George W., and Seppo Honkapohja. 2001. Learning and Expectations in Macroeconomics.
Princeton, NJ: Princeton University Press.
Fama, Eugene F., and Kenneth R. French. 1988. Dividend Yields and Expected Stock Returns. Jour-
nal of Financial Economics 22 (1): 325.
Ferson, Wayne E., Sergei Sarkissian, and Timothy T. Simin. 2003. Spurious Regressions in Financial
Economics? Journal of Finance 58 (4): 13931413.
Greenwood, Robin, and Stefan Nagel. 2009. Inexperienced Investors and Bubbles. Journal of Finan-
cial Economics 93 (2): 23958.
Greenwood, Robin, and Andrei Shleifer. 2014. Expectations of Returns and Expected Returns.
Review of Financial Studies 27 (3): 71446.
Hirota, Shinichi, and Shyam Sunder. 2007. Price Bubbles Sans Dividend Anchors: Evidence from
Laboratory Stock Markets. Journal of Economic Dynamics and Control 31 (6): 18751909.
Hirschleifer, David, Jun Li, and Jianfeng Yu. 2015. Asset Pricing in Production Economics with
Extrapolative Expectations. Journal of Monetary Economics 76: 87106.
Hong, Harrison, Jos Scheinkman, and Wei Xiong. 2006. Asset Float and Speculative Bubbles. Jour-
nal of Finance 61 (3): 10731117.
Lansing, Kevin J. 2010. Rational and Near-Rational Bubbles without Drift. Economic Journal 120
(549): 114974.
Lucas, Robert E., Jr. 1978. Asset Prices in an Exchange Economy. Econometrica 46 (6): 142645.
MacKinnon, James G., and Antony A. Smith, Jr. 1998. Approximate Bias Correction in Economet-
rics. Journal of Econometrics 85 (2): 20530.
Malmendier, Ulrike, and Stefan Nagel. 2011. Depression Babies: Do Macroeconomic Experiences
Affect Risk Taking? Quarterly Journal of Economics 126 (1): 373416.
Marcet, Albert, and Juan P. Nicolini. 2003. Recurrent Hyperinflations and Learning. American Eco-
nomic Review 93 (5): 147698.
Marcet, Albert, and Thomas J. Sargent. 1989. Convergence of Least Squares Learning Mechanisms
in Self-Referential Linear Stochastic Models. Journal of Economic Theory 48 (2): 33768.
Marcet, Albert, and Thomas J. Sargent. 1992. The Convergence of Vector Auto-Regressions to Ratio-
nal Expectations Equilibria. In Macroeconomics: A Strategic Survey, edited by Alessandro Vercelli
and Nicola Dimitri, 13964. Oxford, UK: Oxford University Press.
Marriott, F. H. C., and J. A. Pope. 1954. Bias in the Estimation of Autocorrelations. Biometrika 41
(34): 390402.
Ofek, Eli, and Matthew Richardson. 2003. DotCom Mania: The Rise and Fall of Internet Stock
Prices. Journal of Finance 58 (3): 111337.
Rendahl, Pontus. 2015. Inequality Constraint and Euler Equation-Based Solution Methods. Eco-
nomic Journal 125 (585): 111035.
Roy, Roch. 1989. Asymptotic Covariance Structure of Serial Correlations in Multivariate Time
Series. Biometrika 76 (4): 82427.
Sargent, Thomas J., Noah Williams, and Tao Zha. 2009. The Conquest of South American Inflation.
Journal of Political Economy 117 (2): 21156.
Shiller, Robert J. 1981. Do Stock Prices Move Too Much to Be Justified by Subsequent Changes in
Dividends? American Economic Review 71 (3): 42136.
Shleifer, Andrei, and Robert W. Vishny. 1997. The Limits of Arbitrage. Journal of Finance 52 (1):
3555.
Stambaugh, Robert F. 1999. Predictive Regressions. Journal of Financial Economics 54 (3): 375
421.
Timmermann, Allan G. 1993. How Learning in Financial Markets Generates Excess Volatility and
Predictability in Stock Prices. Quarterly Journal of Economics 108 (4): 113545.
Vissing-Jorgensen, Annette. 2004. Perspectives on Behavioral Finance: Does Irrationality Disap-
pear with Wealth? Evidence from Expectations and Actions. In NBER Macroeconomics Annual
2003, Vol. 18, edited by Mark Gertler and Kenneth Rogoff, 139208. Chicago: University of Chi-
cago Press.
West, Mike, and Jeff Harrison. 1997. Bayesian Forecasting and Dynamic Models. Berlin: Springer.

Adam, K. Et Al. (AER) - Stock Prices Booms and Expected Capital Gains PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Adam, K. Et Al. (AER) - Stock Prices Booms and Expected Capital Gains PDF

Uploaded by

Copyright:

Available Formats

American Economic Review 2017, 107(8): 23522408

Stock Price Booms and Expected Capital Gains

By Klaus Adam, Albert Marcet, and Johannes Beutel*

Investors subjective capital gains expectations are a key element

Bull-markets are born on pessimism, grow on skepticism, mature on

overly pessimistic beliefs at market troughs. Furthermore, it suggests that inves-

Figure 1. Quarterly PD Ratio of the S&P 500, 1946:I2012:I

II. Stock Prices and Stock Price Expectations

Expected return (p.q.)

A. Survey Expectations and the PD Ratio

As pointed out by Vissing-Jorgensen (2004), Greenwood and Shleifer (2014),

B. RE Test with Small Sample Adjustments

(1,Pt /Dt). The operator E

Assuming that the measurement error

Let c NT denote the OLS estimate of c Nwith T observations.

error Nt =Rt, t+NEt(Rt,

Proposition 1: Assume the process { , Nt ,Pt /Dt}is station-

(i) Under the null hypothesis of rational expectations

is a consistent estimate of var(cNcN).23

(ii) Suppose in addition that

for (1,1) , where (utN + Nt ,u Nt , PD

tend to get E (c T cT )<0in small samples, even if in fact c =c .

t+1, t )=cov( t+1

The bias E (T )in equation (9) is approximately given by E (T )

Table 1ARational Expectations Test, H 0: c=c

Survey average Survey median

Panel B. Dow Jones, real price growth

C. Additional Small Sample Adjustments

with given constants A Nand C N. We allow U Nt to be serially correlated by specifying

for t N(0, 2 ),independent of PD

As a result, the regression coefficients ( a N ,c N) in equation (5) do not sat-

Table 1BRational Expectations Test, H 0: cc

Survey average Survey median

Panel B. Dow Jones, real price growth

Table 1CRational Expectations Test, H 0: cc

Survey average Survey median

Panel B. Dow Jones, real price growth

D. How Models of Learning May Help

Adaptive prediction model (g = 0.02641, m(1946:I) = 0%

log price growth expectations

III. A Simple Asset-Pricing Model

Consider an endowment economy populated by a unit mass of infinitely lived

The Investment Problem.Investor i solves

(Pt +Dt)+Wt for all t0,

where C idenotes consumption, uthe instantaneous utility of the consumer,

(15) lnDt = ln D + lnDt1 + ln D

where D 1denotes gross mean dividend growth and l n D

where 1+is the average consumption-dividend ratio and p [0,1)its quarterly

(ln t ) ( 2 ( 2W) (DW 2W ))

The Agents Underlying Probability Space.Agents hold a set of subjective

The fact that C iand S idepend on price realizations is a consequence of optimal

Parametric Utility Function.To obtain closed-form solutions, we consider in

We furthermore assume that

Existence of a Recursive Solution.Since solving the optimization problem (14)

IV. Rational Expectations Equilibrium

Proposition 2: If agents hold rational expectations and if price expectations

V. Learning about Capital Gains and Internal Rationality

A. Agents Belief System: Benchmark Specification

(26) lnPt+1 lnPt = lnt+1

where t +1denotes a transitory shock to price growth and a persistent price

(27) lnt+1 = lnt + lnt+1 .

B. Efficiency of Stock Prices and Internal Rationality

and risk-neutrality deliver the present value relationship

_v+ g(lnPt lnPt1 +____ )